Lazily compute the FSTs during indexing

fix for rankingScoreThreshold changes the results' ranking

.github/ISSUE_TEMPLATE/sprint_issue.md

@@ -22,6 +22,16 @@ Related product discussion:
 
 <!---If necessary, create a list with technical/product steps-->
 
+### Reminders when modifying the API
+
+- [ ] Update the openAPI file with utoipa:
+  - [ ] If a new module has been introduced, create a new structure deriving [the OpenAPI proc-macro](https://docs.rs/utoipa/latest/utoipa/derive.OpenApi.html) and nest it in the main [openAPI structure](https://github.com/meilisearch/meilisearch/blob/f2185438eed60fa32d25b15480c5ee064f6fba4a/crates/meilisearch/src/routes/mod.rs#L64-L78).
+  - [ ] If a new route has been introduced, add the [path decorator](https://docs.rs/utoipa/latest/utoipa/attr.path.html) to it and add the route at the top of the file in its openAPI structure.
+  - [ ] If a structure which is deserialized or serialized in the API has been introduced or modified, it must derive the [`schema`](https://docs.rs/utoipa/latest/utoipa/macro.schema.html) or the [`IntoParams`](https://docs.rs/utoipa/latest/utoipa/derive.IntoParams.html) proc-macro.
+        If it's a **new** structure you must also add it to the big list of structures [in the main `OpenApi` structure](https://github.com/meilisearch/meilisearch/blob/f2185438eed60fa32d25b15480c5ee064f6fba4a/crates/meilisearch/src/routes/mod.rs#L88).
+  - [ ] Once everything is done, start Meilisearch with the swagger flag: `cargo run --features swagger`, open `http://localhost:7700/scalar` on your browser, and ensure everything works as expected.
+  - For more info, refer to [this presentation](https://pitch.com/v/generating-the-openapi-file-jrn3nh).
+
 ### Reminders when modifying the Setting API
 
 <!--- Special steps to remind when adding a new index setting -->

.github/workflows/bench-manual.yml

@@ -18,7 +18,7 @@ jobs:
         timeout-minutes: 180 # 3h
         steps:
             - uses: actions/checkout@v3
-            - uses: dtolnay/rust-toolchain@1.79
+            - uses: dtolnay/rust-toolchain@1.81
               with:
                 profile: minimal
 

.github/workflows/bench-pr.yml

@@ -55,7 +55,7 @@ jobs:
               reaction-type: "rocket"
               repo-token: ${{ env.GH_TOKEN }}
 
-        - uses: xt0rted/pull-request-comment-branch@v2
+        - uses: xt0rted/pull-request-comment-branch@v3
           id: comment-branch
           with:
             repo_token: ${{ env.GH_TOKEN }}
@@ -66,7 +66,7 @@ jobs:
             fetch-depth: 0 # fetch full history to be able to get main commit sha
             ref: ${{ steps.comment-branch.outputs.head_ref }}
 
-        - uses: dtolnay/rust-toolchain@1.79
+        - uses: dtolnay/rust-toolchain@1.81
           with:
             profile: minimal
 

.github/workflows/bench-push-indexing.yml

@@ -12,7 +12,7 @@ jobs:
         timeout-minutes: 180 # 3h
         steps:
           - uses: actions/checkout@v3
-          - uses: dtolnay/rust-toolchain@1.79
+          - uses: dtolnay/rust-toolchain@1.81
             with:
               profile: minimal
 

.github/workflows/benchmarks-manual.yml

@@ -18,7 +18,7 @@ jobs:
     timeout-minutes: 4320 # 72h
     steps:
       - uses: actions/checkout@v3
-      - uses: dtolnay/rust-toolchain@1.79
+      - uses: dtolnay/rust-toolchain@1.81
         with:
           profile: minimal
 

.github/workflows/benchmarks-pr.yml

@@ -44,7 +44,7 @@ jobs:
             exit 1
           fi
 
-      - uses: dtolnay/rust-toolchain@1.79
+      - uses: dtolnay/rust-toolchain@1.81
         with:
           profile: minimal
 
@@ -56,7 +56,7 @@ jobs:
           reaction-type: "eyes"
           repo-token: ${{ env.GH_TOKEN }}
 
-      - uses: xt0rted/pull-request-comment-branch@v2
+      - uses: xt0rted/pull-request-comment-branch@v3
         id: comment-branch
         with:
           repo_token: ${{ env.GH_TOKEN }}

.github/workflows/benchmarks-push-indexing.yml

@@ -16,7 +16,7 @@ jobs:
     timeout-minutes: 4320 # 72h
     steps:
       - uses: actions/checkout@v3
-      - uses: dtolnay/rust-toolchain@1.79
+      - uses: dtolnay/rust-toolchain@1.81
         with:
           profile: minimal
 

.github/workflows/benchmarks-push-search-geo.yml

@@ -15,7 +15,7 @@ jobs:
     runs-on: benchmarks
     steps:
       - uses: actions/checkout@v3
-      - uses: dtolnay/rust-toolchain@1.79
+      - uses: dtolnay/rust-toolchain@1.81
         with:
           profile: minimal
 

.github/workflows/benchmarks-push-search-songs.yml

@@ -15,7 +15,7 @@ jobs:
     runs-on: benchmarks
     steps:
       - uses: actions/checkout@v3
-      - uses: dtolnay/rust-toolchain@1.79
+      - uses: dtolnay/rust-toolchain@1.81
         with:
           profile: minimal
 

.github/workflows/benchmarks-push-search-wiki.yml

@@ -15,7 +15,7 @@ jobs:
     runs-on: benchmarks
     steps:
       - uses: actions/checkout@v3
-      - uses: dtolnay/rust-toolchain@1.79
+      - uses: dtolnay/rust-toolchain@1.81
         with:
           profile: minimal
 

.github/workflows/check-valid-milestone.yml

@@ -0,0 +1,100 @@
+name: PR Milestone Check
+
+on:
+  pull_request:
+    types: [opened, reopened, edited, synchronize, milestoned, demilestoned]
+    branches:
+      - "main"
+      - "release-v*.*.*"
+
+jobs:
+  check-milestone:
+    name: Check PR Milestone
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v3
+
+      - name: Validate PR milestone
+        uses: actions/github-script@v6
+        with:
+          github-token: ${{ secrets.GITHUB_TOKEN }}
+          script: |
+            // Get PR number directly from the event payload
+            const prNumber = context.payload.pull_request.number;
+
+            // Get PR details
+            const { data: prData } = await github.rest.pulls.get({
+              owner: 'meilisearch',
+              repo: 'meilisearch',
+              pull_number: prNumber
+            });
+
+            // Get base branch name
+            const baseBranch = prData.base.ref;
+            console.log(`Base branch: ${baseBranch}`);
+
+            // Get PR milestone
+            const prMilestone = prData.milestone;
+            if (!prMilestone) {
+              core.setFailed('PR must have a milestone assigned');
+              return;
+            }
+            console.log(`PR milestone: ${prMilestone.title}`);
+
+            // Validate milestone format: vx.y.z
+            const milestoneRegex = /^v\d+\.\d+\.\d+$/;
+            if (!milestoneRegex.test(prMilestone.title)) {
+              core.setFailed(`Milestone "${prMilestone.title}" does not follow the required format vx.y.z`);
+              return;
+            }
+
+            // For main branch PRs, check if the milestone is the highest one
+            if (baseBranch === 'main') {
+              // Get all milestones
+              const { data: milestones } = await github.rest.issues.listMilestones({
+                owner: 'meilisearch',
+                repo: 'meilisearch',
+                state: 'open',
+                sort: 'due_on',
+                direction: 'desc'
+              });
+
+              // Sort milestones by version number (vx.y.z)
+              const sortedMilestones = milestones
+                .filter(m => milestoneRegex.test(m.title))
+                .sort((a, b) => {
+                  const versionA = a.title.substring(1).split('.').map(Number);
+                  const versionB = b.title.substring(1).split('.').map(Number);
+
+                  // Compare major version
+                  if (versionA[0] !== versionB[0]) return versionB[0] - versionA[0];
+                  // Compare minor version
+                  if (versionA[1] !== versionB[1]) return versionB[1] - versionA[1];
+                  // Compare patch version
+                  return versionB[2] - versionA[2];
+                });
+
+              if (sortedMilestones.length === 0) {
+                core.setFailed('No valid milestones found in the repository. Please create at least one milestone with the format vx.y.z');
+                return;
+              }
+
+              const highestMilestone = sortedMilestones[0];
+              console.log(`Highest milestone: ${highestMilestone.title}`);
+
+              if (prMilestone.title !== highestMilestone.title) {
+                core.setFailed(`PRs targeting the main branch must use the highest milestone (${highestMilestone.title}), but this PR uses ${prMilestone.title}`);
+                return;
+              }
+            } else {
+              // For release branches, the milestone should match the branch version
+              const branchVersion = baseBranch.substring(8); // remove 'release-'
+              if (prMilestone.title !== branchVersion) {
+                core.setFailed(`PRs targeting release branch "${baseBranch}" must use the matching milestone "${branchVersion}", but this PR uses "${prMilestone.title}"`);
+                return;
+              }
+            }
+
+            console.log('PR milestone validation passed!');

.github/workflows/flaky-tests.yml

@@ -9,22 +9,22 @@ jobs:
   flaky:
     runs-on: ubuntu-latest
     container:
-      # Use ubuntu-20.04 to compile with glibc 2.28
-      image: ubuntu:20.04
+      # Use ubuntu-22.04 to compile with glibc 2.35
+      image: ubuntu:22.04
     steps:
-    - uses: actions/checkout@v3
-    - name: Install needed dependencies
-      run: |
-        apt-get update && apt-get install -y curl
-        apt-get install build-essential -y
-    - uses: dtolnay/rust-toolchain@1.79
-    - name: Install cargo-flaky
-      run: cargo install cargo-flaky
-    - name: Run cargo flaky in the dumps
-      run: cd crates/dump; cargo flaky -i 100 --release
-    - name: Run cargo flaky in the index-scheduler
-      run: cd crates/index-scheduler; cargo flaky -i 100 --release
-    - name: Run cargo flaky in the auth
-      run: cd crates/meilisearch-auth; cargo flaky -i 100 --release
-    - name: Run cargo flaky in meilisearch
-      run: cd crates/meilisearch; cargo flaky -i 100 --release
+      - uses: actions/checkout@v3
+      - name: Install needed dependencies
+        run: |
+          apt-get update && apt-get install -y curl
+          apt-get install build-essential -y
+      - uses: dtolnay/rust-toolchain@1.81
+      - name: Install cargo-flaky
+        run: cargo install cargo-flaky
+      - name: Run cargo flaky in the dumps
+        run: cd crates/dump; cargo flaky -i 100 --release
+      - name: Run cargo flaky in the index-scheduler
+        run: cd crates/index-scheduler; cargo flaky -i 100 --release
+      - name: Run cargo flaky in the auth
+        run: cd crates/meilisearch-auth; cargo flaky -i 100 --release
+      - name: Run cargo flaky in meilisearch
+        run: cd crates/meilisearch; cargo flaky -i 100 --release

.github/workflows/fuzzer-indexing.yml

@@ -12,7 +12,7 @@ jobs:
     timeout-minutes: 4320 # 72h
     steps:
       - uses: actions/checkout@v3
-      - uses: dtolnay/rust-toolchain@1.79
+      - uses: dtolnay/rust-toolchain@1.81
         with:
           profile: minimal
 

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

@@ -18,28 +18,28 @@ jobs:
     runs-on: ubuntu-latest
     needs: check-version
     container:
-      # Use ubuntu-20.04 to compile with glibc 2.28
-      image: ubuntu:20.04
+      # Use ubuntu-22.04 to compile with glibc 2.35
+      image: ubuntu:22.04
     steps:
-    - name: Install needed dependencies
-      run: |
-        apt-get update && apt-get install -y curl
-        apt-get install build-essential -y
-    - uses: dtolnay/rust-toolchain@1.79
-    - name: Install cargo-deb
-      run: cargo install cargo-deb
-    - uses: actions/checkout@v3
-    - 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@2.7.0
-      with:
-        repo_token: ${{ secrets.MEILI_BOT_GH_PAT }}
-        file: target/debian/meilisearch.deb
-        asset_name: meilisearch.deb
-        tag: ${{ github.ref }}
-    - name: Upload debian pkg to apt repository
-      run: curl -F package=@target/debian/meilisearch.deb https://${{ secrets.GEMFURY_PUSH_TOKEN }}@push.fury.io/meilisearch/
+      - name: Install needed dependencies
+        run: |
+          apt-get update && apt-get install -y curl
+          apt-get install build-essential -y
+      - uses: dtolnay/rust-toolchain@1.81
+      - name: Install cargo-deb
+        run: cargo install cargo-deb
+      - uses: actions/checkout@v3
+      - 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@2.7.0
+        with:
+          repo_token: ${{ secrets.MEILI_BOT_GH_PAT }}
+          file: target/debian/meilisearch.deb
+          asset_name: meilisearch.deb
+          tag: ${{ github.ref }}
+      - name: Upload debian pkg to apt repository
+        run: curl -F package=@target/debian/meilisearch.deb https://${{ secrets.GEMFURY_PUSH_TOKEN }}@push.fury.io/meilisearch/
 
   homebrew:
     name: Bump Homebrew formula

.github/workflows/publish-binaries.yml

@@ -3,7 +3,7 @@ name: Publish binaries to GitHub release
 on:
   workflow_dispatch:
   schedule:
-    - cron: '0 2 * * *' # Every day at 2:00am
+    - cron: "0 2 * * *" # Every day at 2:00am
   release:
     types: [published]
 
@@ -37,26 +37,26 @@ jobs:
     runs-on: ubuntu-latest
     needs: check-version
     container:
-      # Use ubuntu-20.04 to compile with glibc 2.28
-      image: ubuntu:20.04
+      # Use ubuntu-22.04 to compile with glibc 2.35
+      image: ubuntu:22.04
     steps:
-    - uses: actions/checkout@v3
-    - name: Install needed dependencies
-      run: |
-        apt-get update && apt-get install -y curl
-        apt-get install build-essential -y
-    - uses: dtolnay/rust-toolchain@1.79
-    - name: Build
-      run: cargo build --release --locked
-    # No need to upload binaries for dry run (cron)
-    - name: Upload binaries to release
-      if: github.event_name == 'release'
-      uses: svenstaro/upload-release-action@2.7.0
-      with:
-        repo_token: ${{ secrets.MEILI_BOT_GH_PAT }}
-        file: target/release/meilisearch
-        asset_name: meilisearch-linux-amd64
-        tag: ${{ github.ref }}
+      - uses: actions/checkout@v3
+      - name: Install needed dependencies
+        run: |
+          apt-get update && apt-get install -y curl
+          apt-get install build-essential -y
+      - uses: dtolnay/rust-toolchain@1.81
+      - name: Build
+        run: cargo build --release --locked
+      # No need to upload binaries for dry run (cron)
+      - name: Upload binaries to release
+        if: github.event_name == 'release'
+        uses: svenstaro/upload-release-action@2.7.0
+        with:
+          repo_token: ${{ secrets.MEILI_BOT_GH_PAT }}
+          file: target/release/meilisearch
+          asset_name: meilisearch-linux-amd64
+          tag: ${{ github.ref }}
 
   publish-macos-windows:
     name: Publish binary for ${{ matrix.os }}
@@ -74,19 +74,19 @@ jobs:
             artifact_name: meilisearch.exe
             asset_name: meilisearch-windows-amd64.exe
     steps:
-    - uses: actions/checkout@v3
-    - uses: dtolnay/rust-toolchain@1.79
-    - name: Build
-      run: cargo build --release --locked
-    # No need to upload binaries for dry run (cron)
-    - name: Upload binaries to release
-      if: github.event_name == 'release'
-      uses: svenstaro/upload-release-action@2.7.0
-      with:
-        repo_token: ${{ secrets.MEILI_BOT_GH_PAT }}
-        file: target/release/${{ matrix.artifact_name }}
-        asset_name: ${{ matrix.asset_name }}
-        tag: ${{ github.ref }}
+      - uses: actions/checkout@v3
+      - uses: dtolnay/rust-toolchain@1.81
+      - name: Build
+        run: cargo build --release --locked
+      # No need to upload binaries for dry run (cron)
+      - name: Upload binaries to release
+        if: github.event_name == 'release'
+        uses: svenstaro/upload-release-action@2.7.0
+        with:
+          repo_token: ${{ secrets.MEILI_BOT_GH_PAT }}
+          file: target/release/${{ matrix.artifact_name }}
+          asset_name: ${{ matrix.asset_name }}
+          tag: ${{ github.ref }}
 
   publish-macos-apple-silicon:
     name: Publish binary for macOS silicon
@@ -101,7 +101,7 @@ jobs:
       - name: Checkout repository
         uses: actions/checkout@v3
       - name: Installing Rust toolchain
-        uses: dtolnay/rust-toolchain@1.79
+        uses: dtolnay/rust-toolchain@1.81
         with:
           profile: minimal
           target: ${{ matrix.target }}
@@ -127,8 +127,8 @@ jobs:
     env:
       DEBIAN_FRONTEND: noninteractive
     container:
-      # Use ubuntu-20.04 to compile with glibc 2.28
-      image: ubuntu:20.04
+      # Use ubuntu-22.04 to compile with glibc 2.35
+      image: ubuntu:22.04
     strategy:
       matrix:
         include:
@@ -148,7 +148,7 @@ jobs:
           add-apt-repository "deb [arch=$(dpkg --print-architecture)] https://download.docker.com/linux/ubuntu $(lsb_release -cs) stable"
           apt-get update -y && apt-get install -y docker-ce
       - name: Installing Rust toolchain
-        uses: dtolnay/rust-toolchain@1.79
+        uses: dtolnay/rust-toolchain@1.81
         with:
           profile: minimal
           target: ${{ matrix.target }}

.github/workflows/sdks-tests.yml

@@ -52,7 +52,7 @@ jobs:
       - name: Setup .NET Core
         uses: actions/setup-dotnet@v4
         with:
-          dotnet-version: "6.0.x"
+          dotnet-version: "8.0.x"
       - name: Install dependencies
         run: dotnet restore
       - name: Build

.github/workflows/test-suite.yml

@@ -4,13 +4,9 @@ on:
   workflow_dispatch:
   schedule:
     # Everyday at 5:00am
-    - cron: '0 5 * * *'
+    - cron: "0 5 * * *"
   pull_request:
-  push:
-    # trying and staging branches are for Bors config
-    branches:
-      - trying
-      - staging
+  merge_group:
 
 env:
   CARGO_TERM_COLOR: always
@@ -19,21 +15,21 @@ env:
 
 jobs:
   test-linux:
-    name: Tests on ubuntu-20.04
+    name: Tests on ubuntu-22.04
     runs-on: ubuntu-latest
     container:
-      # Use ubuntu-20.04 to compile with glibc 2.28
-      image: ubuntu:20.04
+      # Use ubuntu-22.04 to compile with glibc 2.35
+      image: ubuntu:22.04
     steps:
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@v4
       - name: Install needed dependencies
         run: |
           apt-get update && apt-get install -y curl
           apt-get install build-essential -y
       - name: Setup test with Rust stable
-        uses: dtolnay/rust-toolchain@1.79
+        uses: dtolnay/rust-toolchain@1.81
       - name: Cache dependencies
-        uses: Swatinem/rust-cache@v2.7.5
+        uses: Swatinem/rust-cache@v2.7.7
       - name: Run cargo check without any default features
         uses: actions-rs/cargo@v1
         with:
@@ -55,8 +51,8 @@ jobs:
     steps:
       - uses: actions/checkout@v3
       - name: Cache dependencies
-        uses: Swatinem/rust-cache@v2.7.5
-      - uses: dtolnay/rust-toolchain@1.79
+        uses: Swatinem/rust-cache@v2.7.7
+      - uses: dtolnay/rust-toolchain@1.81
       - name: Run cargo check without any default features
         uses: actions-rs/cargo@v1
         with:
@@ -72,8 +68,8 @@ jobs:
     name: Tests almost all features
     runs-on: ubuntu-latest
     container:
-      # Use ubuntu-20.04 to compile with glibc 2.28
-      image: ubuntu:20.04
+      # Use ubuntu-22.04 to compile with glibc 2.35
+      image: ubuntu:22.04
     if: github.event_name == 'schedule' || github.event_name == 'workflow_dispatch'
     steps:
       - uses: actions/checkout@v3
@@ -81,19 +77,51 @@ jobs:
         run: |
           apt-get update
           apt-get install --assume-yes build-essential curl
-      - uses: dtolnay/rust-toolchain@1.79
+      - uses: dtolnay/rust-toolchain@1.81
       - name: Run cargo build with almost all features
         run: |
-          cargo build --workspace --locked --release --features "$(cargo xtask list-features --exclude-feature cuda)"
+          cargo build --workspace --locked --release --features "$(cargo xtask list-features --exclude-feature cuda,test-ollama)"
       - name: Run cargo test with almost all features
         run: |
-          cargo test --workspace --locked --release --features "$(cargo xtask list-features --exclude-feature cuda)"
+          cargo test --workspace --locked --release --features "$(cargo xtask list-features --exclude-feature cuda,test-ollama)"
+
+  ollama-ubuntu:
+    name: Test with Ollama
+    runs-on: ubuntu-latest
+    env:
+      MEILI_TEST_OLLAMA_SERVER: "http://localhost:11434"
+    steps:
+      - uses: actions/checkout@v1
+      - name: Install Ollama
+        run: |
+          curl -fsSL https://ollama.com/install.sh | sudo -E sh
+      - name: Start serving
+        run: |
+          # Run it in the background, there is no way to daemonise at the moment
+          ollama serve &
+
+          # A short pause is required before the HTTP port is opened
+          sleep 5
+
+          # This endpoint blocks until ready
+          time curl -i http://localhost:11434
+
+      - name: Pull nomic-embed-text & all-minilm
+        run: |
+          ollama pull nomic-embed-text
+          ollama pull all-minilm
+
+      - name: Run cargo test
+        uses: actions-rs/cargo@v1
+        with:
+          command: test
+          args: --locked --release --all --features test-ollama ollama
 
   test-disabled-tokenization:
     name: Test disabled tokenization
     runs-on: ubuntu-latest
     container:
-      image: ubuntu:20.04
+      image: ubuntu:22.04
     if: github.event_name == 'schedule' || github.event_name == 'workflow_dispatch'
     steps:
       - uses: actions/checkout@v3
@@ -101,7 +129,7 @@ jobs:
         run: |
           apt-get update
           apt-get install --assume-yes build-essential curl
-      - uses: dtolnay/rust-toolchain@1.79
+      - uses: dtolnay/rust-toolchain@1.81
       - name: Run cargo tree without default features and check lindera is not present
         run: |
           if cargo tree -f '{p} {f}' -e normal --no-default-features | grep -qz lindera; then
@@ -117,17 +145,17 @@ jobs:
     name: Run tests in debug
     runs-on: ubuntu-latest
     container:
-      # Use ubuntu-20.04 to compile with glibc 2.28
-      image: ubuntu:20.04
+      # Use ubuntu-22.04 to compile with glibc 2.35
+      image: ubuntu:22.04
     steps:
       - uses: actions/checkout@v3
       - name: Install needed dependencies
         run: |
           apt-get update && apt-get install -y curl
           apt-get install build-essential -y
-      - uses: dtolnay/rust-toolchain@1.79
+      - uses: dtolnay/rust-toolchain@1.81
       - name: Cache dependencies
-        uses: Swatinem/rust-cache@v2.7.5
+        uses: Swatinem/rust-cache@v2.7.7
       - name: Run tests in debug
         uses: actions-rs/cargo@v1
         with:
@@ -139,12 +167,12 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v3
-      - uses: dtolnay/rust-toolchain@1.79
+      - uses: dtolnay/rust-toolchain@1.81
         with:
           profile: minimal
           components: clippy
       - name: Cache dependencies
-        uses: Swatinem/rust-cache@v2.7.5
+        uses: Swatinem/rust-cache@v2.7.7
       - name: Run cargo clippy
         uses: actions-rs/cargo@v1
         with:
@@ -156,14 +184,14 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v3
-      - uses: dtolnay/rust-toolchain@1.79
+      - uses: dtolnay/rust-toolchain@1.81
         with:
           profile: minimal
           toolchain: nightly-2024-07-09
           override: true
           components: rustfmt
       - name: Cache dependencies
-        uses: Swatinem/rust-cache@v2.7.5
+        uses: Swatinem/rust-cache@v2.7.7
       - 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

.github/workflows/update-cargo-toml-version.yml

@@ -18,7 +18,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v3
-      - uses: dtolnay/rust-toolchain@1.79
+      - uses: dtolnay/rust-toolchain@1.81
         with:
           profile: minimal
       - name: Install sd

.gitignore

@@ -10,6 +10,7 @@
 /dumps
 /bench
 /_xtask_benchmark.ms
+/benchmarks
 
 # Snapshots
 ## ... large

BENCHMARKS.md

@@ -48,6 +48,27 @@ cargo xtask bench --no-dashboard -- workloads/my_workload_1.json workloads/my_wo
 
 For processing the results, look at [Looking at benchmark results/Without dashboard](#without-dashboard).
 
+#### Sending a workload by hand
+
+Sometimes you want to visualize the metrics of a worlkoad that comes from a custom report.
+It is not quite easy to trick the benchboard in thinking that your report is legitimate but here are the commands you can run to upload your firefox report on a running benchboard.
+
+```bash
+# Name this hostname whatever you want
+echo '{ "hostname": "the-best-place" }' | xh PUT 'http://127.0.0.1:9001/api/v1/machine'
+
+# You'll receive an UUID from this command that we will call $invocation_uuid
+echo '{ "commit": { "sha1": "1234567", "commit_date": "2024-09-05 12:00:12.0 +00:00:00", "message": "A cool message" }, "machine_hostname": "the-best-place", "max_workloads": 1 }' | xh PUT 'http://127.0.0.1:9001/api/v1/invocation'
+
+# Just use UUID from the previous command
+# and you'll receive another UUID that we will call $workload_uuid
+echo '{ "invocation_uuid": "$invocation_uuid", "name": "toto", "max_runs": 1 }' | xh PUT 'http://127.0.0.1:9001/api/v1/workload'
+
+# And now use your $workload_uuid and the content of your firefox report
+# but don't forget to convert your firefox report from JSONLines into an object
+echo '{ "workload_uuid": "$workload_uuid", "data": $REPORT_JSON_DATA }' | xh PUT 'http://127.0.0.1:9001/api/v1/run'
+```
+
 ### In CI
 
 We have dedicated runners to run workloads on CI. Currently, there are three ways of running the CI:

CONTRIBUTING.md

@@ -95,6 +95,11 @@ Meilisearch follows the [cargo xtask](https://github.com/matklad/cargo-xtask) wo
 
 Run `cargo xtask --help` from the root of the repository to find out what is available.
 
+#### Update the openAPI file if the API changed
+
+To update the openAPI file in the code, see [sprint_issue.md](https://github.com/meilisearch/meilisearch/blob/main/.github/ISSUE_TEMPLATE/sprint_issue.md#reminders-when-modifying-the-api).
+If you want to update the openAPI file on the [open-api repository](https://github.com/meilisearch/open-api), see [update-openapi-issue.md](https://github.com/meilisearch/engine-team/blob/main/issue-templates/update-openapi-issue.md).
+
 ### Logging
 
 Meilisearch uses [`tracing`](https://lib.rs/crates/tracing) for logging purposes. Tracing logs are structured and can be displayed as JSON to the end user, so prefer passing arguments as fields rather than interpolating them in the message.
@@ -145,7 +150,7 @@ Some notes on GitHub PRs:
 - The PR title should be accurate and descriptive of the changes.
 - [Convert your PR as a draft](https://help.github.com/en/github/collaborating-with-issues-and-pull-requests/changing-the-stage-of-a-pull-request) if your changes are a work in progress: no one will review it until you pass your PR as ready for review.<br>
   The draft PRs are recommended when you want to show that you are working on something and make your work visible.
-- The branch related to the PR must be **up-to-date with `main`** before merging. Fortunately, this project uses [Bors](https://github.com/bors-ng/bors-ng) to automatically enforce this requirement without the PR author having to rebase manually.
+- The branch related to the PR must be **up-to-date with `main`** before merging. Fortunately, this project uses [GitHub Merge Queues](https://github.blog/news-insights/product-news/github-merge-queue-is-generally-available/) to automatically enforce this requirement without the PR author having to rebase manually.
 
 ## Release Process (for internal team only)
 
@@ -153,8 +158,7 @@ Meilisearch tools follow the [Semantic Versioning Convention](https://semver.org
 
 ### Automation to rebase and Merge the PRs
 
-This project integrates a bot that helps us manage pull requests merging.<br>
-_[Read more about this](https://github.com/meilisearch/integration-guides/blob/main/resources/bors.md)._
+This project uses GitHub Merge Queues that helps us manage pull requests merging.
 
 ### How to Publish a new Release
 

Cargo.toml

@@ -22,7 +22,7 @@ members = [
 ]
 
 [workspace.package]
-version = "1.12.0"
+version = "1.14.0"
 authors = [
     "Quentin de Quelen <quentin@dequelen.me>",
     "Clément Renault <clement@meilisearch.com>",
@@ -36,6 +36,12 @@ license = "MIT"
 [profile.release]
 codegen-units = 1
 
+# We now compile heed without the NDEBUG define for better performance.
+# However, we still enable debug assertions for a better detection of
+# disk corruption on the cloud or in OSS.
+[profile.release.package.heed]
+debug-assertions = true
+
 [profile.dev.package.flate2]
 opt-level = 3
 
@@ -43,6 +49,3 @@ opt-level = 3
 opt-level = 3
 [profile.dev.package.roaring]
 opt-level = 3
-
-[patch.crates-io]
-roaring = { git = "https://github.com/RoaringBitmap/roaring-rs", branch = "clone-iter-slice" }

Dockerfile

@@ -1,5 +1,5 @@
 # Compile
-FROM    rust:1.79.0-alpine3.20 AS compiler
+FROM    rust:1.81.0-alpine3.20 AS compiler
 
 RUN     apk add -q --no-cache build-base openssl-dev
 

LICENSE

@@ -1,6 +1,6 @@
 MIT License
 
-Copyright (c) 2019-2024 Meili SAS
+Copyright (c) 2019-2025 Meili SAS
 
 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal

README.md

@@ -20,7 +20,7 @@
 <p align="center">
   <a href="https://deps.rs/repo/github/meilisearch/meilisearch"><img src="https://deps.rs/repo/github/meilisearch/meilisearch/status.svg" alt="Dependency status"></a>
   <a href="https://github.com/meilisearch/meilisearch/blob/main/LICENSE"><img src="https://img.shields.io/badge/license-MIT-informational" alt="License"></a>
-  <a href="https://ms-bors.herokuapp.com/repositories/52"><img src="https://bors.tech/images/badge_small.svg" alt="Bors enabled"></a>
+  <a href="https://github.com/meilisearch/meilisearch/queue"><img alt="Merge Queues enabled" src="https://img.shields.io/badge/Merge_Queues-enabled-%2357cf60?logo=github"></a>
 </p>
 
 <p align="center">⚡ A lightning-fast search engine that fits effortlessly into your apps, websites, and workflow 🔍</p>
@@ -58,6 +58,7 @@ See the list of all our example apps in our [demos repository](https://github.co
 - **[Multi-Tenancy](https://www.meilisearch.com/docs/learn/security/multitenancy_tenant_tokens?utm_campaign=oss&utm_source=github&utm_medium=meilisearch&utm_content=features):** personalize search results for any number of application tenants
 - **Highly Customizable:** customize Meilisearch to your specific needs or use our out-of-the-box and hassle-free presets
 - **[RESTful API](https://www.meilisearch.com/docs/reference/api/overview?utm_campaign=oss&utm_source=github&utm_medium=meilisearch&utm_content=features):** integrate Meilisearch in your technical stack with our plugins and SDKs
+- **AI-ready:** works out of the box with [langchain](https://www.meilisearch.com/with/langchain) and the [model context protocol](https://github.com/meilisearch/meilisearch-mcp)
 - **Easy to install, deploy, and maintain**
 
 ## 📖 Documentation

assets/grafana-dashboard.json

@@ -1403,6 +1403,104 @@
       "title": "Number of tasks by indexes",
       "type": "timeseries"
     },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "${DS_PROMETHEUS}"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "color": {
+            "mode": "palette-classic"
+          },
+          "custom": {
+            "axisBorderShow": false,
+            "axisCenteredZero": false,
+            "axisColorMode": "text",
+            "axisLabel": "",
+            "axisPlacement": "auto",
+            "barAlignment": 0,
+            "drawStyle": "line",
+            "fillOpacity": 15,
+            "gradientMode": "none",
+            "hideFrom": {
+              "legend": false,
+              "tooltip": false,
+              "viz": false
+            },
+            "insertNulls": false,
+            "lineInterpolation": "linear",
+            "lineWidth": 1,
+            "pointSize": 5,
+            "scaleDistribution": {
+              "type": "linear"
+            },
+            "showPoints": "never",
+            "spanNulls": false,
+            "stacking": {
+              "group": "A",
+              "mode": "none"
+            },
+            "thresholdsStyle": {
+              "mode": "off"
+            }
+          },
+          "mappings": [],
+          "thresholds": {
+            "mode": "absolute",
+            "steps": [
+              {
+                "color": "green"
+              },
+              {
+                "color": "red",
+                "value": 80
+              }
+            ]
+          },
+          "unit": "none"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 11,
+        "w": 12,
+        "x": 12,
+        "y": 51
+      },
+      "id": 29,
+      "interval": "5s",
+      "options": {
+        "legend": {
+          "calcs": [],
+          "displayMode": "list",
+          "placement": "right",
+          "showLegend": true
+        },
+        "tooltip": {
+          "mode": "single",
+          "sort": "none"
+        }
+      },
+      "pluginVersion": "8.1.4",
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "${DS_PROMETHEUS}"
+          },
+          "editorMode": "builder",
+          "exemplar": true,
+          "expr": "meilisearch_task_queue_latency_seconds{instance=\"$instance\", job=\"$job\"}",
+          "interval": "",
+          "legendFormat": "{{value}} ",
+          "range": true,
+          "refId": "A"
+        }
+      ],
+      "title": "Task queue latency",
+      "type": "timeseries"
+    },
     {
       "collapsed": true,
       "datasource": {

bors.toml

@@ -1,11 +0,0 @@
-status = [
-    'Tests on ubuntu-20.04',
-    'Tests on macos-13',
-    'Tests on windows-2022',
-    'Run Clippy',
-    'Run Rustfmt',
-    'Run tests in debug',
-]
-pr_status = ['Milestone Check']
-# 3 hours timeout
-timeout-sec = 10800

crates/benchmarks/Cargo.toml

@@ -11,27 +11,27 @@ edition.workspace = true
 license.workspace = true
 
 [dependencies]
-anyhow = "1.0.86"
+anyhow = "1.0.95"
 bumpalo = "3.16.0"
-csv = "1.3.0"
+csv = "1.3.1"
 memmap2 = "0.9.5"
 milli = { path = "../milli" }
 mimalloc = { version = "0.1.43", default-features = false }
-serde_json = { version = "1.0.120", features = ["preserve_order"] }
-tempfile = "3.14.0"
+serde_json = { version = "1.0.135", features = ["preserve_order"] }
+tempfile = "3.15.0"
 
 [dev-dependencies]
 criterion = { version = "0.5.1", features = ["html_reports"] }
 rand = "0.8.5"
 rand_chacha = "0.3.1"
-roaring = "0.10.6"
+roaring = "0.10.10"
 
 [build-dependencies]
-anyhow = "1.0.86"
-bytes = "1.6.0"
+anyhow = "1.0.95"
+bytes = "1.9.0"
 convert_case = "0.6.0"
-flate2 = "1.0.30"
-reqwest = { version = "0.12.5", features = ["blocking", "rustls-tls"], default-features = false }
+flate2 = "1.0.35"
+reqwest = { version = "0.12.12", features = ["blocking", "rustls-tls"], default-features = false }
 
 [features]
 default = ["milli/all-tokenizations"]

crates/benchmarks/benches/search_geo.rs

@@ -2,9 +2,10 @@ mod datasets_paths;
 mod utils;
 
 use criterion::{criterion_group, criterion_main};
-use milli::update::Settings;
+use milli::{update::Settings, FilterableAttributesRule};
 use utils::Conf;
 
+#[cfg(not(windows))]
 #[global_allocator]
 static ALLOC: mimalloc::MiMalloc = mimalloc::MiMalloc;
 
@@ -20,8 +21,10 @@ fn base_conf(builder: &mut Settings) {
         ["name", "alternatenames", "elevation"].iter().map(|s| s.to_string()).collect();
     builder.set_searchable_fields(searchable_fields);
 
-    let filterable_fields =
-        ["_geo", "population", "elevation"].iter().map(|s| s.to_string()).collect();
+    let filterable_fields = ["_geo", "population", "elevation"]
+        .iter()
+        .map(|s| FilterableAttributesRule::Field(s.to_string()))
+        .collect();
     builder.set_filterable_fields(filterable_fields);
 
     let sortable_fields =

crates/benchmarks/benches/search_songs.rs

@@ -2,9 +2,10 @@ mod datasets_paths;
 mod utils;
 
 use criterion::{criterion_group, criterion_main};
-use milli::update::Settings;
+use milli::{update::Settings, FilterableAttributesRule};
 use utils::Conf;
 
+#[cfg(not(windows))]
 #[global_allocator]
 static ALLOC: mimalloc::MiMalloc = mimalloc::MiMalloc;
 
@@ -21,7 +22,7 @@ fn base_conf(builder: &mut Settings) {
 
     let faceted_fields = ["released-timestamp", "duration-float", "genre", "country", "artist"]
         .iter()
-        .map(|s| s.to_string())
+        .map(|s| FilterableAttributesRule::Field(s.to_string()))
         .collect();
     builder.set_filterable_fields(faceted_fields);
 }

crates/benchmarks/benches/search_wiki.rs

@@ -5,6 +5,7 @@ use criterion::{criterion_group, criterion_main};
 use milli::update::Settings;
 use utils::Conf;
 
+#[cfg(not(windows))]
 #[global_allocator]
 static ALLOC: mimalloc::MiMalloc = mimalloc::MiMalloc;
 

crates/benchmarks/benches/utils.rs

@@ -10,8 +10,9 @@ use bumpalo::Bump;
 use criterion::BenchmarkId;
 use memmap2::Mmap;
 use milli::heed::EnvOpenOptions;
+use milli::progress::Progress;
 use milli::update::new::indexer;
-use milli::update::{IndexDocumentsMethod, IndexerConfig, Settings};
+use milli::update::{IndexerConfig, Settings};
 use milli::vector::EmbeddingConfigs;
 use milli::{Criterion, Filter, Index, Object, TermsMatchingStrategy};
 use serde_json::Value;
@@ -64,10 +65,11 @@ pub fn base_setup(conf: &Conf) -> Index {
     }
     create_dir_all(conf.database_name).unwrap();
 
-    let mut options = EnvOpenOptions::new();
+    let options = EnvOpenOptions::new();
+    let mut options = options.read_txn_without_tls();
     options.map_size(100 * 1024 * 1024 * 1024); // 100 GB
     options.max_readers(100);
-    let index = Index::new(options, conf.database_name).unwrap();
+    let index = Index::new(options, conf.database_name, true).unwrap();
 
     let config = IndexerConfig::default();
     let mut wtxn = index.write_txn().unwrap();
@@ -98,8 +100,8 @@ pub fn base_setup(conf: &Conf) -> Index {
     let mut new_fields_ids_map = db_fields_ids_map.clone();
 
     let documents = documents_from(conf.dataset, conf.dataset_format);
-    let mut indexer = indexer::DocumentOperation::new(IndexDocumentsMethod::ReplaceDocuments);
-    indexer.add_documents(&documents).unwrap();
+    let mut indexer = indexer::DocumentOperation::new();
+    indexer.replace_documents(&documents).unwrap();
 
     let indexer_alloc = Bump::new();
     let (document_changes, _operation_stats, primary_key) = indexer
@@ -110,13 +112,14 @@ pub fn base_setup(conf: &Conf) -> Index {
             None,
             &mut new_fields_ids_map,
             &|| false,
-            &|_progress| (),
+            Progress::default(),
         )
         .unwrap();
 
     indexer::index(
         &mut wtxn,
         &index,
+        &milli::ThreadPoolNoAbortBuilder::new().build().unwrap(),
         config.grenad_parameters(),
         &db_fields_ids_map,
         new_fields_ids_map,
@@ -124,7 +127,7 @@ pub fn base_setup(conf: &Conf) -> Index {
         &document_changes,
         EmbeddingConfigs::default(),
         &|| false,
-        &|_| (),
+        &Progress::default(),
     )
     .unwrap();
 

crates/build-info/Cargo.toml

@@ -11,8 +11,8 @@ license.workspace = true
 # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
 
 [dependencies]
-time = { version = "0.3.36", features = ["parsing"] }
+time = { version = "0.3.37", features = ["parsing"] }
 
 [build-dependencies]
-anyhow = "1.0.86"
-vergen-git2 = "1.0.0"
+anyhow = "1.0.95"
+vergen-git2 = "1.0.2"

crates/dump/Cargo.toml

@@ -11,21 +11,21 @@ readme.workspace = true
 license.workspace = true
 
 [dependencies]
-anyhow = "1.0.86"
-flate2 = "1.0.30"
-http = "1.1.0"
+anyhow = "1.0.95"
+flate2 = "1.0.35"
+http = "1.2.0"
 meilisearch-types = { path = "../meilisearch-types" }
-once_cell = "1.19.0"
-regex = "1.10.5"
-roaring = { version = "0.10.6", features = ["serde"] }
-serde = { version = "1.0.204", features = ["derive"] }
-serde_json = { version = "1.0.120", features = ["preserve_order"] }
-tar = "0.4.41"
-tempfile = "3.10.1"
-thiserror = "1.0.61"
-time = { version = "0.3.36", features = ["serde-well-known", "formatting", "parsing", "macros"] }
-tracing = "0.1.40"
-uuid = { version = "1.10.0", features = ["serde", "v4"] }
+once_cell = "1.20.2"
+regex = "1.11.1"
+roaring = { version = "0.10.10", features = ["serde"] }
+serde = { version = "1.0.217", features = ["derive"] }
+serde_json = { version = "1.0.135", features = ["preserve_order"] }
+tar = "0.4.43"
+tempfile = "3.15.0"
+thiserror = "2.0.9"
+time = { version = "0.3.37", features = ["serde-well-known", "formatting", "parsing", "macros"] }
+tracing = "0.1.41"
+uuid = { version = "1.11.0", features = ["serde", "v4"] }
 
 [dev-dependencies]
 big_s = "1.0.2"

crates/dump/README.md

@@ -10,8 +10,10 @@ dump
 ├── instance-uid.uuid
 ├── keys.jsonl
 ├── metadata.json
-└── tasks
-    ├── update_files
-    │   └── [task_id].jsonl
+├── tasks
+│   ├── update_files
+│   │   └── [task_id].jsonl
+│   └── queue.jsonl
+└── batches
     └── queue.jsonl
-```
+```

crates/dump/src/lib.rs

@@ -141,6 +141,9 @@ pub enum KindDump {
         instance_uid: Option<InstanceUid>,
     },
     SnapshotCreation,
+    UpgradeDatabase {
+        from: (u32, u32, u32),
+    },
 }
 
 impl From<Task> for TaskDump {
@@ -210,6 +213,9 @@ impl From<KindWithContent> for KindDump {
                 KindDump::DumpCreation { keys, instance_uid }
             }
             KindWithContent::SnapshotCreation => KindDump::SnapshotCreation,
+            KindWithContent::UpgradeDatabase { from: version } => {
+                KindDump::UpgradeDatabase { from: version }
+            }
         }
     }
 }
@@ -222,14 +228,16 @@ pub(crate) mod test {
 
     use big_s::S;
     use maplit::{btreemap, btreeset};
+    use meilisearch_types::batches::{Batch, BatchEnqueuedAt, BatchStats};
     use meilisearch_types::facet_values_sort::FacetValuesSort;
-    use meilisearch_types::features::RuntimeTogglableFeatures;
+    use meilisearch_types::features::{Network, Remote, RuntimeTogglableFeatures};
     use meilisearch_types::index_uid_pattern::IndexUidPattern;
     use meilisearch_types::keys::{Action, Key};
-    use meilisearch_types::milli;
     use meilisearch_types::milli::update::Setting;
+    use meilisearch_types::milli::{self, FilterableAttributesRule};
     use meilisearch_types::settings::{Checked, FacetingSettings, Settings};
-    use meilisearch_types::tasks::{Details, Status};
+    use meilisearch_types::task_view::DetailsView;
+    use meilisearch_types::tasks::{Details, Kind, Status};
     use serde_json::{json, Map, Value};
     use time::macros::datetime;
     use uuid::Uuid;
@@ -271,7 +279,10 @@ pub(crate) mod test {
         let settings = Settings {
             displayed_attributes: Setting::Set(vec![S("race"), S("name")]).into(),
             searchable_attributes: Setting::Set(vec![S("name"), S("race")]).into(),
-            filterable_attributes: Setting::Set(btreeset! { S("race"), S("age") }),
+            filterable_attributes: Setting::Set(vec![
+                FilterableAttributesRule::Field(S("race")),
+                FilterableAttributesRule::Field(S("age")),
+            ]),
             sortable_attributes: Setting::Set(btreeset! { S("age") }),
             ranking_rules: Setting::NotSet,
             stop_words: Setting::NotSet,
@@ -299,6 +310,33 @@ pub(crate) mod test {
         settings.check()
     }
 
+    pub fn create_test_batches() -> Vec<Batch> {
+        vec![Batch {
+            uid: 0,
+            details: DetailsView {
+                received_documents: Some(12),
+                indexed_documents: Some(Some(10)),
+                ..DetailsView::default()
+            },
+            progress: None,
+            stats: BatchStats {
+                total_nb_tasks: 1,
+                status: maplit::btreemap! { Status::Succeeded => 1 },
+                types: maplit::btreemap! { Kind::DocumentAdditionOrUpdate => 1 },
+                index_uids: maplit::btreemap! { "doggo".to_string() => 1 },
+                progress_trace: Default::default(),
+                write_channel_congestion: None,
+                internal_database_sizes: Default::default(),
+            },
+            enqueued_at: Some(BatchEnqueuedAt {
+                earliest: datetime!(2022-11-11 0:00 UTC),
+                oldest: datetime!(2022-11-11 0:00 UTC),
+            }),
+            started_at: datetime!(2022-11-20 0:00 UTC),
+            finished_at: Some(datetime!(2022-11-21 0:00 UTC)),
+        }]
+    }
+
     pub fn create_test_tasks() -> Vec<(TaskDump, Option<Vec<Document>>)> {
         vec![
             (
@@ -421,6 +459,15 @@ pub(crate) mod test {
         index.flush().unwrap();
         index.settings(&settings).unwrap();
 
+        // ========== pushing the batch queue
+        let batches = create_test_batches();
+
+        let mut batch_queue = dump.create_batches_queue().unwrap();
+        for batch in &batches {
+            batch_queue.push_batch(batch).unwrap();
+        }
+        batch_queue.flush().unwrap();
+
         // ========== pushing the task queue
         let tasks = create_test_tasks();
 
@@ -449,6 +496,10 @@ pub(crate) mod test {
 
         dump.create_experimental_features(features).unwrap();
 
+        // ========== network
+        let network = create_test_network();
+        dump.create_network(network).unwrap();
+
         // create the dump
         let mut file = tempfile::tempfile().unwrap();
         dump.persist_to(&mut file).unwrap();
@@ -458,7 +509,14 @@ pub(crate) mod test {
     }
 
     fn create_test_features() -> RuntimeTogglableFeatures {
-        RuntimeTogglableFeatures { vector_store: true, ..Default::default() }
+        RuntimeTogglableFeatures::default()
+    }
+
+    fn create_test_network() -> Network {
+        Network {
+            local: Some("myself".to_string()),
+            remotes: maplit::btreemap! {"other".to_string() => Remote { url: "http://test".to_string(), search_api_key: Some("apiKey".to_string()) }},
+        }
     }
 
     #[test]
@@ -509,5 +567,9 @@ pub(crate) mod test {
         // ==== checking the features
         let expected = create_test_features();
         assert_eq!(dump.features().unwrap().unwrap(), expected);
+
+        // ==== checking the network
+        let expected = create_test_network();
+        assert_eq!(&expected, dump.network().unwrap().unwrap());
     }
 }

crates/dump/src/reader/compat/v5_to_v6.rs

@@ -196,6 +196,10 @@ impl CompatV5ToV6 {
     pub fn features(&self) -> Result<Option<v6::RuntimeTogglableFeatures>> {
         Ok(None)
     }
+
+    pub fn network(&self) -> Result<Option<&v6::Network>> {
+        Ok(None)
+    }
 }
 
 pub enum CompatIndexV5ToV6 {
@@ -318,7 +322,16 @@ impl<T> From<v5::Settings<T>> for v6::Settings<v6::Unchecked> {
         v6::Settings {
             displayed_attributes: v6::Setting::from(settings.displayed_attributes).into(),
             searchable_attributes: v6::Setting::from(settings.searchable_attributes).into(),
-            filterable_attributes: settings.filterable_attributes.into(),
+            filterable_attributes: match settings.filterable_attributes {
+                v5::settings::Setting::Set(filterable_attributes) => v6::Setting::Set(
+                    filterable_attributes
+                        .into_iter()
+                        .map(v6::FilterableAttributesRule::Field)
+                        .collect(),
+                ),
+                v5::settings::Setting::Reset => v6::Setting::Reset,
+                v5::settings::Setting::NotSet => v6::Setting::NotSet,
+            },
             sortable_attributes: settings.sortable_attributes.into(),
             ranking_rules: {
                 match settings.ranking_rules {

crates/dump/src/reader/mod.rs

@@ -23,6 +23,7 @@ mod v6;
 pub type Document = serde_json::Map<String, serde_json::Value>;
 pub type UpdateFile = dyn Iterator<Item = Result<Document>>;
 
+#[allow(clippy::large_enum_variant)]
 pub enum DumpReader {
     Current(V6Reader),
     Compat(CompatV5ToV6),
@@ -101,6 +102,13 @@ impl DumpReader {
         }
     }
 
+    pub fn batches(&mut self) -> Result<Box<dyn Iterator<Item = Result<v6::Batch>> + '_>> {
+        match self {
+            DumpReader::Current(current) => Ok(current.batches()),
+            DumpReader::Compat(_compat) => Ok(Box::new(std::iter::empty())),
+        }
+    }
+
     pub fn keys(&mut self) -> Result<Box<dyn Iterator<Item = Result<v6::Key>> + '_>> {
         match self {
             DumpReader::Current(current) => Ok(current.keys()),
@@ -114,6 +122,13 @@ impl DumpReader {
             DumpReader::Compat(compat) => compat.features(),
         }
     }
+
+    pub fn network(&self) -> Result<Option<&v6::Network>> {
+        match self {
+            DumpReader::Current(current) => Ok(current.network()),
+            DumpReader::Compat(compat) => compat.network(),
+        }
+    }
 }
 
 impl From<V6Reader> for DumpReader {
@@ -219,6 +234,10 @@ pub(crate) mod test {
         insta::assert_snapshot!(dump.date().unwrap(), @"2024-05-16 15:51:34.151044 +00:00:00");
         insta::assert_debug_snapshot!(dump.instance_uid().unwrap(), @"None");
 
+        // batches didn't exists at the time
+        let batches = dump.batches().unwrap().collect::<Result<Vec<_>>>().unwrap();
+        meili_snap::snapshot!(meili_snap::json_string!(batches), @"[]");
+
         // tasks
         let tasks = dump.tasks().unwrap().collect::<Result<Vec<_>>>().unwrap();
         let (tasks, update_files): (Vec<_>, Vec<_>) = tasks.into_iter().unzip();
@@ -327,10 +346,8 @@ pub(crate) mod test {
             }
         }
 
-        assert_eq!(
-            dump.features().unwrap().unwrap(),
-            RuntimeTogglableFeatures { vector_store: true, ..Default::default() }
-        );
+        assert_eq!(dump.features().unwrap().unwrap(), RuntimeTogglableFeatures::default());
+        assert_eq!(dump.network().unwrap(), None);
     }
 
     #[test]
@@ -342,6 +359,10 @@ pub(crate) mod test {
         insta::assert_snapshot!(dump.date().unwrap(), @"2023-07-06 7:10:27.21958 +00:00:00");
         insta::assert_debug_snapshot!(dump.instance_uid().unwrap(), @"None");
 
+        // batches didn't exists at the time
+        let batches = dump.batches().unwrap().collect::<Result<Vec<_>>>().unwrap();
+        meili_snap::snapshot!(meili_snap::json_string!(batches), @"[]");
+
         // tasks
         let tasks = dump.tasks().unwrap().collect::<Result<Vec<_>>>().unwrap();
         let (tasks, update_files): (Vec<_>, Vec<_>) = tasks.into_iter().unzip();
@@ -373,10 +394,28 @@ pub(crate) mod test {
 
         assert_eq!(test.documents().unwrap().count(), 1);
 
-        assert_eq!(
-            dump.features().unwrap().unwrap(),
-            RuntimeTogglableFeatures { vector_store: true, ..Default::default() }
-        );
+        assert_eq!(dump.features().unwrap().unwrap(), RuntimeTogglableFeatures::default());
+    }
+
+    #[test]
+    fn import_dump_v6_network() {
+        let dump = File::open("tests/assets/v6-with-network.dump").unwrap();
+        let dump = DumpReader::open(dump).unwrap();
+
+        // top level infos
+        insta::assert_snapshot!(dump.date().unwrap(), @"2025-01-29 15:45:32.738676 +00:00:00");
+        insta::assert_debug_snapshot!(dump.instance_uid().unwrap(), @"None");
+
+        // network
+
+        let network = dump.network().unwrap().unwrap();
+        insta::assert_snapshot!(network.local.as_ref().unwrap(), @"ms-0");
+        insta::assert_snapshot!(network.remotes.get("ms-0").as_ref().unwrap().url, @"http://localhost:7700");
+        insta::assert_snapshot!(network.remotes.get("ms-0").as_ref().unwrap().search_api_key.is_none(), @"true");
+        insta::assert_snapshot!(network.remotes.get("ms-1").as_ref().unwrap().url, @"http://localhost:7701");
+        insta::assert_snapshot!(network.remotes.get("ms-1").as_ref().unwrap().search_api_key.is_none(), @"true");
+        insta::assert_snapshot!(network.remotes.get("ms-2").as_ref().unwrap().url, @"http://ms-5679.example.meilisearch.io");
+        insta::assert_snapshot!(network.remotes.get("ms-2").as_ref().unwrap().search_api_key.as_ref().unwrap(), @"foo");
     }
 
     #[test]
@@ -388,6 +427,10 @@ pub(crate) mod test {
         insta::assert_snapshot!(dump.date().unwrap(), @"2022-10-04 15:55:10.344982459 +00:00:00");
         insta::assert_snapshot!(dump.instance_uid().unwrap().unwrap(), @"9e15e977-f2ae-4761-943f-1eaf75fd736d");
 
+        // batches didn't exists at the time
+        let batches = dump.batches().unwrap().collect::<Result<Vec<_>>>().unwrap();
+        meili_snap::snapshot!(meili_snap::json_string!(batches), @"[]");
+
         // tasks
         let tasks = dump.tasks().unwrap().collect::<Result<Vec<_>>>().unwrap();
         let (tasks, update_files): (Vec<_>, Vec<_>) = tasks.into_iter().unzip();
@@ -468,6 +511,10 @@ pub(crate) mod test {
         insta::assert_snapshot!(dump.date().unwrap(), @"2022-10-06 12:53:49.131989609 +00:00:00");
         insta::assert_snapshot!(dump.instance_uid().unwrap().unwrap(), @"9e15e977-f2ae-4761-943f-1eaf75fd736d");
 
+        // batches didn't exists at the time
+        let batches = dump.batches().unwrap().collect::<Result<Vec<_>>>().unwrap();
+        meili_snap::snapshot!(meili_snap::json_string!(batches), @"[]");
+
         // tasks
         let tasks = dump.tasks().unwrap().collect::<Result<Vec<_>>>().unwrap();
         let (tasks, update_files): (Vec<_>, Vec<_>) = tasks.into_iter().unzip();
@@ -545,6 +592,10 @@ pub(crate) mod test {
         insta::assert_snapshot!(dump.date().unwrap(), @"2022-10-07 11:39:03.709153554 +00:00:00");
         assert_eq!(dump.instance_uid().unwrap(), None);
 
+        // batches didn't exists at the time
+        let batches = dump.batches().unwrap().collect::<Result<Vec<_>>>().unwrap();
+        meili_snap::snapshot!(meili_snap::json_string!(batches), @"[]");
+
         // tasks
         let tasks = dump.tasks().unwrap().collect::<Result<Vec<_>>>().unwrap();
         let (tasks, update_files): (Vec<_>, Vec<_>) = tasks.into_iter().unzip();
@@ -638,6 +689,10 @@ pub(crate) mod test {
         insta::assert_snapshot!(dump.date().unwrap(), @"2022-10-09 20:27:59.904096267 +00:00:00");
         assert_eq!(dump.instance_uid().unwrap(), None);
 
+        // batches didn't exists at the time
+        let batches = dump.batches().unwrap().collect::<Result<Vec<_>>>().unwrap();
+        meili_snap::snapshot!(meili_snap::json_string!(batches), @"[]");
+
         // tasks
         let tasks = dump.tasks().unwrap().collect::<Result<Vec<_>>>().unwrap();
         let (tasks, update_files): (Vec<_>, Vec<_>) = tasks.into_iter().unzip();
@@ -731,6 +786,10 @@ pub(crate) mod test {
         insta::assert_snapshot!(dump.date().unwrap(), @"2023-01-30 16:26:09.247261 +00:00:00");
         assert_eq!(dump.instance_uid().unwrap(), None);
 
+        // batches didn't exists at the time
+        let batches = dump.batches().unwrap().collect::<Result<Vec<_>>>().unwrap();
+        meili_snap::snapshot!(meili_snap::json_string!(batches), @"[]");
+
         // tasks
         let tasks = dump.tasks().unwrap().collect::<Result<Vec<_>>>().unwrap();
         let (tasks, update_files): (Vec<_>, Vec<_>) = tasks.into_iter().unzip();
@@ -807,6 +866,10 @@ pub(crate) mod test {
         assert_eq!(dump.date(), None);
         assert_eq!(dump.instance_uid().unwrap(), None);
 
+        // batches didn't exists at the time
+        let batches = dump.batches().unwrap().collect::<Result<Vec<_>>>().unwrap();
+        meili_snap::snapshot!(meili_snap::json_string!(batches), @"[]");
+
         // tasks
         let tasks = dump.tasks().unwrap().collect::<Result<Vec<_>>>().unwrap();
         let (tasks, update_files): (Vec<_>, Vec<_>) = tasks.into_iter().unzip();

crates/dump/src/reader/snapshots/dump__reader__test__import_dump_v6_with_vectors-5.snap

@@ -1,5 +1,5 @@
 ---
-source: dump/src/reader/mod.rs
+source: crates/dump/src/reader/mod.rs
 expression: vector_index.settings().unwrap()
 ---
 {
@@ -49,6 +49,7 @@ expression: vector_index.settings().unwrap()
       "source": "huggingFace",
       "model": "BAAI/bge-base-en-v1.5",
       "revision": "617ca489d9e86b49b8167676d8220688b99db36e",
+      "pooling": "forceMean",
       "documentTemplate": "{% for field in fields %} {{ field.name }}: {{ field.value }}\n{% endfor %}"
     }
   },

crates/dump/src/reader/v6/mod.rs

@@ -3,6 +3,7 @@ use std::io::{BufRead, BufReader, ErrorKind};
 use std::path::Path;
 
 pub use meilisearch_types::milli;
+use meilisearch_types::milli::vector::hf::OverridePooling;
 use tempfile::TempDir;
 use time::OffsetDateTime;
 use tracing::debug;
@@ -18,8 +19,10 @@ pub type Checked = meilisearch_types::settings::Checked;
 pub type Unchecked = meilisearch_types::settings::Unchecked;
 
 pub type Task = crate::TaskDump;
+pub type Batch = meilisearch_types::batches::Batch;
 pub type Key = meilisearch_types::keys::Key;
 pub type RuntimeTogglableFeatures = meilisearch_types::features::RuntimeTogglableFeatures;
+pub type Network = meilisearch_types::features::Network;
 
 // ===== Other types to clarify the code of the compat module
 // everything related to the tasks
@@ -43,13 +46,17 @@ pub type ResponseError = meilisearch_types::error::ResponseError;
 pub type Code = meilisearch_types::error::Code;
 pub type RankingRuleView = meilisearch_types::settings::RankingRuleView;
 
+pub type FilterableAttributesRule = meilisearch_types::milli::FilterableAttributesRule;
+
 pub struct V6Reader {
     dump: TempDir,
     instance_uid: Option<Uuid>,
     metadata: Metadata,
     tasks: BufReader<File>,
+    batches: Option<BufReader<File>>,
     keys: BufReader<File>,
     features: Option<RuntimeTogglableFeatures>,
+    network: Option<Network>,
 }
 
 impl V6Reader {
@@ -77,13 +84,38 @@ impl V6Reader {
         } else {
             None
         };
+        let batches = match File::open(dump.path().join("batches").join("queue.jsonl")) {
+            Ok(file) => Some(BufReader::new(file)),
+            // The batch file was only introduced during the v1.13, anything prior to that won't have batches
+            Err(err) if err.kind() == ErrorKind::NotFound => None,
+            Err(e) => return Err(e.into()),
+        };
+
+        let network_file = match fs::read(dump.path().join("network.json")) {
+            Ok(network_file) => Some(network_file),
+            Err(error) => match error.kind() {
+                // Allows the file to be missing, this will only result in all experimental features disabled.
+                ErrorKind::NotFound => {
+                    debug!("`network.json` not found in dump");
+                    None
+                }
+                _ => return Err(error.into()),
+            },
+        };
+        let network = if let Some(network_file) = network_file {
+            Some(serde_json::from_reader(&*network_file)?)
+        } else {
+            None
+        };
 
         Ok(V6Reader {
             metadata: serde_json::from_reader(&*meta_file)?,
             instance_uid,
             tasks: BufReader::new(File::open(dump.path().join("tasks").join("queue.jsonl"))?),
+            batches,
             keys: BufReader::new(File::open(dump.path().join("keys.jsonl"))?),
             features,
+            network,
             dump,
         })
     }
@@ -124,7 +156,7 @@ impl V6Reader {
         &mut self,
     ) -> Box<dyn Iterator<Item = Result<(Task, Option<Box<super::UpdateFile>>)>> + '_> {
         Box::new((&mut self.tasks).lines().map(|line| -> Result<_> {
-            let task: Task = serde_json::from_str(&line?).unwrap();
+            let task: Task = serde_json::from_str(&line?)?;
 
             let update_file_path = self
                 .dump
@@ -136,8 +168,7 @@ impl V6Reader {
             if update_file_path.exists() {
                 Ok((
                     task,
-                    Some(Box::new(UpdateFile::new(&update_file_path).unwrap())
-                        as Box<super::UpdateFile>),
+                    Some(Box::new(UpdateFile::new(&update_file_path)?) as Box<super::UpdateFile>),
                 ))
             } else {
                 Ok((task, None))
@@ -145,6 +176,16 @@ impl V6Reader {
         }))
     }
 
+    pub fn batches(&mut self) -> Box<dyn Iterator<Item = Result<Batch>> + '_> {
+        match self.batches.as_mut() {
+            Some(batches) => Box::new((batches).lines().map(|line| -> Result<_> {
+                let batch = serde_json::from_str(&line?)?;
+                Ok(batch)
+            })),
+            None => Box::new(std::iter::empty()) as Box<dyn Iterator<Item = Result<Batch>> + '_>,
+        }
+    }
+
     pub fn keys(&mut self) -> Box<dyn Iterator<Item = Result<Key>> + '_> {
         Box::new(
             (&mut self.keys).lines().map(|line| -> Result<_> { Ok(serde_json::from_str(&line?)?) }),
@@ -154,6 +195,10 @@ impl V6Reader {
     pub fn features(&self) -> Option<RuntimeTogglableFeatures> {
         self.features
     }
+
+    pub fn network(&self) -> Option<&Network> {
+        self.network.as_ref()
+    }
 }
 
 pub struct UpdateFile {
@@ -210,7 +255,29 @@ impl V6IndexReader {
     }
 
     pub fn settings(&mut self) -> Result<Settings<Checked>> {
-        let settings: Settings<Unchecked> = serde_json::from_reader(&mut self.settings)?;
+        let mut settings: Settings<Unchecked> = serde_json::from_reader(&mut self.settings)?;
+        patch_embedders(&mut settings);
         Ok(settings.check())
     }
 }
+
+fn patch_embedders(settings: &mut Settings<Unchecked>) {
+    if let Setting::Set(embedders) = &mut settings.embedders {
+        for settings in embedders.values_mut() {
+            let Setting::Set(settings) = &mut settings.inner else {
+                continue;
+            };
+            if settings.source != Setting::Set(milli::vector::settings::EmbedderSource::HuggingFace)
+            {
+                continue;
+            }
+            settings.pooling = match settings.pooling {
+                Setting::Set(pooling) => Setting::Set(pooling),
+                // if the pooling for a hugging face embedder is not set, force it to `forceMean`
+                // for backward compatibility with v1.13
+                // dumps created in v1.14 and up will have the setting set for hugging face embedders
+                Setting::Reset | Setting::NotSet => Setting::Set(OverridePooling::ForceMean),
+            };
+        }
+    }
+}

crates/dump/src/writer.rs

@@ -4,7 +4,8 @@ use std::path::PathBuf;
 
 use flate2::write::GzEncoder;
 use flate2::Compression;
-use meilisearch_types::features::RuntimeTogglableFeatures;
+use meilisearch_types::batches::Batch;
+use meilisearch_types::features::{Network, RuntimeTogglableFeatures};
 use meilisearch_types::keys::Key;
 use meilisearch_types::settings::{Checked, Settings};
 use serde_json::{Map, Value};
@@ -54,6 +55,10 @@ impl DumpWriter {
         TaskWriter::new(self.dir.path().join("tasks"))
     }
 
+    pub fn create_batches_queue(&self) -> Result<BatchWriter> {
+        BatchWriter::new(self.dir.path().join("batches"))
+    }
+
     pub fn create_experimental_features(&self, features: RuntimeTogglableFeatures) -> Result<()> {
         Ok(std::fs::write(
             self.dir.path().join("experimental-features.json"),
@@ -61,6 +66,10 @@ impl DumpWriter {
         )?)
     }
 
+    pub fn create_network(&self, network: Network) -> Result<()> {
+        Ok(std::fs::write(self.dir.path().join("network.json"), serde_json::to_string(&network)?)?)
+    }
+
     pub fn persist_to(self, mut writer: impl Write) -> Result<()> {
         let gz_encoder = GzEncoder::new(&mut writer, Compression::default());
         let mut tar_encoder = tar::Builder::new(gz_encoder);
@@ -84,7 +93,7 @@ impl KeyWriter {
     }
 
     pub fn push_key(&mut self, key: &Key) -> Result<()> {
-        self.keys.write_all(&serde_json::to_vec(key)?)?;
+        serde_json::to_writer(&mut self.keys, &key)?;
         self.keys.write_all(b"\n")?;
         Ok(())
     }
@@ -114,7 +123,7 @@ impl TaskWriter {
     /// Pushes tasks in the dump.
     /// If the tasks has an associated `update_file` it'll use the `task_id` as its name.
     pub fn push_task(&mut self, task: &TaskDump) -> Result<UpdateFile> {
-        self.queue.write_all(&serde_json::to_vec(task)?)?;
+        serde_json::to_writer(&mut self.queue, &task)?;
         self.queue.write_all(b"\n")?;
 
         Ok(UpdateFile::new(self.update_files.join(format!("{}.jsonl", task.uid))))
@@ -126,6 +135,30 @@ impl TaskWriter {
     }
 }
 
+pub struct BatchWriter {
+    queue: BufWriter<File>,
+}
+
+impl BatchWriter {
+    pub(crate) fn new(path: PathBuf) -> Result<Self> {
+        std::fs::create_dir(&path)?;
+        let queue = File::create(path.join("queue.jsonl"))?;
+        Ok(BatchWriter { queue: BufWriter::new(queue) })
+    }
+
+    /// Pushes batches in the dump.
+    pub fn push_batch(&mut self, batch: &Batch) -> Result<()> {
+        serde_json::to_writer(&mut self.queue, &batch)?;
+        self.queue.write_all(b"\n")?;
+        Ok(())
+    }
+
+    pub fn flush(mut self) -> Result<()> {
+        self.queue.flush()?;
+        Ok(())
+    }
+}
+
 pub struct UpdateFile {
     path: PathBuf,
     writer: Option<BufWriter<File>>,
@@ -137,8 +170,8 @@ impl UpdateFile {
     }
 
     pub fn push_document(&mut self, document: &Document) -> Result<()> {
-        if let Some(writer) = self.writer.as_mut() {
-            writer.write_all(&serde_json::to_vec(document)?)?;
+        if let Some(mut writer) = self.writer.as_mut() {
+            serde_json::to_writer(&mut writer, &document)?;
             writer.write_all(b"\n")?;
         } else {
             let file = File::create(&self.path).unwrap();
@@ -205,8 +238,8 @@ pub(crate) mod test {
     use super::*;
     use crate::reader::Document;
     use crate::test::{
-        create_test_api_keys, create_test_documents, create_test_dump, create_test_instance_uid,
-        create_test_settings, create_test_tasks,
+        create_test_api_keys, create_test_batches, create_test_documents, create_test_dump,
+        create_test_instance_uid, create_test_settings, create_test_tasks,
     };
 
     fn create_directory_hierarchy(dir: &Path) -> String {
@@ -281,8 +314,10 @@ pub(crate) mod test {
         let dump_path = dump.path();
 
         // ==== checking global file hierarchy (we want to be sure there isn't too many files or too few)
-        insta::assert_snapshot!(create_directory_hierarchy(dump_path), @r###"
+        insta::assert_snapshot!(create_directory_hierarchy(dump_path), @r"
         .
+        ├---- batches/
+        │    └---- queue.jsonl
         ├---- indexes/
         │    └---- doggos/
         │    │    ├---- documents.jsonl
@@ -295,8 +330,9 @@ pub(crate) mod test {
         ├---- experimental-features.json
         ├---- instance_uid.uuid
         ├---- keys.jsonl
-        └---- metadata.json
-        "###);
+        ├---- metadata.json
+        └---- network.json
+        ");
 
         // ==== checking the top level infos
         let metadata = fs::read_to_string(dump_path.join("metadata.json")).unwrap();
@@ -349,6 +385,16 @@ pub(crate) mod test {
             }
         }
 
+        // ==== checking the batch queue
+        let batches_queue = fs::read_to_string(dump_path.join("batches/queue.jsonl")).unwrap();
+        for (batch, expected) in batches_queue.lines().zip(create_test_batches()) {
+            let mut batch = serde_json::from_str::<Batch>(batch).unwrap();
+            if batch.details.settings == Some(Box::new(Settings::<Unchecked>::default())) {
+                batch.details.settings = None;
+            }
+            assert_eq!(batch, expected, "{batch:#?}{expected:#?}");
+        }
+
         // ==== checking the keys
         let keys = fs::read_to_string(dump_path.join("keys.jsonl")).unwrap();
         for (key, expected) in keys.lines().zip(create_test_api_keys()) {

crates/file-store/Cargo.toml

@@ -11,7 +11,7 @@ edition.workspace = true
 license.workspace = true
 
 [dependencies]
-tempfile = "3.10.1"
-thiserror = "1.0.61"
-tracing = "0.1.40"
-uuid = { version = "1.10.0", features = ["serde", "v4"] }
+tempfile = "3.15.0"
+thiserror = "2.0.9"
+tracing = "0.1.41"
+uuid = { version = "1.11.0", features = ["serde", "v4"] }

crates/file-store/src/lib.rs

@@ -136,6 +136,14 @@ pub struct File {
 }
 
 impl File {
+    pub fn from_parts(path: PathBuf, file: Option<NamedTempFile>) -> Self {
+        Self { path, file }
+    }
+
+    pub fn into_parts(self) -> (PathBuf, Option<NamedTempFile>) {
+        (self.path, self.file)
+    }
+
     pub fn dry_file() -> Result<Self> {
         Ok(Self { path: PathBuf::new(), file: None })
     }

crates/filter-parser/Cargo.toml

@@ -17,4 +17,5 @@ nom_locate = "4.2.0"
 unescaper = "0.1.5"
 
 [dev-dependencies]
-insta = "1.39.0"
+# fixed version due to format breakages in v1.40
+insta = "=1.39.0"

crates/filter-parser/src/condition.rs

@@ -30,6 +30,25 @@ pub enum Condition<'a> {
     StartsWith { keyword: Token<'a>, word: Token<'a> },
 }
 
+impl Condition<'_> {
+    pub fn operator(&self) -> &str {
+        match self {
+            Condition::GreaterThan(_) => ">",
+            Condition::GreaterThanOrEqual(_) => ">=",
+            Condition::Equal(_) => "=",
+            Condition::NotEqual(_) => "!=",
+            Condition::Null => "IS NULL",
+            Condition::Empty => "IS EMPTY",
+            Condition::Exists => "EXISTS",
+            Condition::LowerThan(_) => "<",
+            Condition::LowerThanOrEqual(_) => "<=",
+            Condition::Between { .. } => "TO",
+            Condition::Contains { .. } => "CONTAINS",
+            Condition::StartsWith { .. } => "STARTS WITH",
+        }
+    }
+}
+
 /// condition      = value ("==" | ">" ...) value
 pub fn parse_condition(input: Span) -> IResult<FilterCondition> {
     let operator = alt((tag("<="), tag(">="), tag("!="), tag("<"), tag(">"), tag("=")));

crates/filter-parser/src/lib.rs

@@ -179,6 +179,26 @@ impl<'a> FilterCondition<'a> {
         }
     }
 
+    pub fn fids(&self, depth: usize) -> Box<dyn Iterator<Item = &Token> + '_> {
+        if depth == 0 {
+            return Box::new(std::iter::empty());
+        }
+        match self {
+            FilterCondition::Condition { fid, .. } | FilterCondition::In { fid, .. } => {
+                Box::new(std::iter::once(fid))
+            }
+            FilterCondition::Not(filter) => {
+                let depth = depth.saturating_sub(1);
+                filter.fids(depth)
+            }
+            FilterCondition::And(subfilters) | FilterCondition::Or(subfilters) => {
+                let depth = depth.saturating_sub(1);
+                Box::new(subfilters.iter().flat_map(move |f| f.fids(depth)))
+            }
+            _ => Box::new(std::iter::empty()),
+        }
+    }
+
     /// Returns the first token found at the specified depth, `None` if no token at this depth.
     pub fn token_at_depth(&self, depth: usize) -> Option<&Token> {
         match self {
@@ -978,6 +998,43 @@ pub mod tests {
         assert!(filter.token_at_depth(3).is_none());
     }
 
+    #[test]
+    fn fids() {
+        let filter = Fc::parse("field = value").unwrap().unwrap();
+        let fids: Vec<_> = filter.fids(MAX_FILTER_DEPTH).collect();
+        assert_eq!(fids.len(), 1);
+        assert_eq!(fids[0].value(), "field");
+
+        let filter = Fc::parse("field IN [1, 2, 3]").unwrap().unwrap();
+        let fids: Vec<_> = filter.fids(MAX_FILTER_DEPTH).collect();
+        assert_eq!(fids.len(), 1);
+        assert_eq!(fids[0].value(), "field");
+
+        let filter = Fc::parse("field != value").unwrap().unwrap();
+        let fids: Vec<_> = filter.fids(MAX_FILTER_DEPTH).collect();
+        assert_eq!(fids.len(), 1);
+        assert_eq!(fids[0].value(), "field");
+
+        let filter = Fc::parse("field1 = value1 AND field2 = value2").unwrap().unwrap();
+        let fids: Vec<_> = filter.fids(MAX_FILTER_DEPTH).collect();
+        assert_eq!(fids.len(), 2);
+        assert!(fids[0].value() == "field1");
+        assert!(fids[1].value() == "field2");
+
+        let filter = Fc::parse("field1 = value1 OR field2 = value2").unwrap().unwrap();
+        let fids: Vec<_> = filter.fids(MAX_FILTER_DEPTH).collect();
+        assert_eq!(fids.len(), 2);
+        assert!(fids[0].value() == "field1");
+        assert!(fids[1].value() == "field2");
+
+        let depth = 2;
+        let filter =
+            Fc::parse("field1 = value1 AND (field2 = value2 OR field3 = value3)").unwrap().unwrap();
+        let fids: Vec<_> = filter.fids(depth).collect();
+        assert_eq!(fids.len(), 1);
+        assert_eq!(fids[0].value(), "field1");
+    }
+
     #[test]
     fn token_from_str() {
         let s = "test string that should not be parsed";

crates/fuzzers/Cargo.toml

@@ -11,12 +11,12 @@ edition.workspace = true
 license.workspace = true
 
 [dependencies]
-arbitrary = { version = "1.3.2", features = ["derive"] }
+arbitrary = { version = "1.4.1", features = ["derive"] }
 bumpalo = "3.16.0"
-clap = { version = "4.5.9", features = ["derive"] }
+clap = { version = "4.5.24", features = ["derive"] }
 either = "1.13.0"
-fastrand = "2.1.0"
+fastrand = "2.3.0"
 milli = { path = "../milli" }
-serde = { version = "1.0.204", features = ["derive"] }
-serde_json = { version = "1.0.120", features = ["preserve_order"] }
-tempfile = "3.10.1"
+serde = { version = "1.0.217", features = ["derive"] }
+serde_json = { version = "1.0.135", features = ["preserve_order"] }
+tempfile = "3.15.0"

crates/fuzzers/src/bin/fuzz-indexing.rs

@@ -10,8 +10,9 @@ use either::Either;
 use fuzzers::Operation;
 use milli::documents::mmap_from_objects;
 use milli::heed::EnvOpenOptions;
+use milli::progress::Progress;
 use milli::update::new::indexer;
-use milli::update::{IndexDocumentsMethod, IndexerConfig};
+use milli::update::IndexerConfig;
 use milli::vector::EmbeddingConfigs;
 use milli::Index;
 use serde_json::Value;
@@ -56,13 +57,14 @@ fn main() {
         let opt = opt.clone();
 
         let handle = std::thread::spawn(move || {
-            let mut options = EnvOpenOptions::new();
+            let options = EnvOpenOptions::new();
+            let mut options = options.read_txn_without_tls();
             options.map_size(1024 * 1024 * 1024 * 1024);
             let tempdir = match opt.path {
                 Some(path) => TempDir::new_in(path).unwrap(),
                 None => TempDir::new().unwrap(),
             };
-            let index = Index::new(options, tempdir.path()).unwrap();
+            let index = Index::new(options, tempdir.path(), true).unwrap();
             let indexer_config = IndexerConfig::default();
 
             std::thread::scope(|s| {
@@ -88,9 +90,7 @@ fn main() {
 
                             let indexer_alloc = Bump::new();
                             let embedders = EmbeddingConfigs::default();
-                            let mut indexer = indexer::DocumentOperation::new(
-                                IndexDocumentsMethod::ReplaceDocuments,
-                            );
+                            let mut indexer = indexer::DocumentOperation::new();
 
                             let mut operations = Vec::new();
                             for op in batch.0 {
@@ -114,7 +114,7 @@ fn main() {
                             for op in &operations {
                                 match op {
                                     Either::Left(documents) => {
-                                        indexer.add_documents(documents).unwrap()
+                                        indexer.replace_documents(documents).unwrap()
                                     }
                                     Either::Right(ids) => indexer.delete_documents(ids),
                                 }
@@ -128,13 +128,14 @@ fn main() {
                                     None,
                                     &mut new_fields_ids_map,
                                     &|| false,
-                                    &|_progress| (),
+                                    Progress::default(),
                                 )
                                 .unwrap();
 
                             indexer::index(
                                 &mut wtxn,
                                 &index,
+                                &milli::ThreadPoolNoAbortBuilder::new().build().unwrap(),
                                 indexer_config.grenad_parameters(),
                                 &db_fields_ids_map,
                                 new_fields_ids_map,
@@ -142,7 +143,7 @@ fn main() {
                                 &document_changes,
                                 embedders,
                                 &|| false,
-                                &|_| (),
+                                &Progress::default(),
                             )
                             .unwrap();
 

crates/index-scheduler/Cargo.toml

@@ -11,41 +11,44 @@ edition.workspace = true
 license.workspace = true
 
 [dependencies]
-anyhow = "1.0.86"
+anyhow = "1.0.95"
 bincode = "1.3.3"
-csv = "1.3.0"
-derive_builder = "0.20.0"
+byte-unit = "5.1.6"
+bumpalo = "3.16.0"
+bumparaw-collections = "0.1.4"
+convert_case = "0.6.0"
+csv = "1.3.1"
+derive_builder = "0.20.2"
 dump = { path = "../dump" }
 enum-iterator = "2.1.0"
 file-store = { path = "../file-store" }
-flate2 = "1.0.30"
+flate2 = "1.0.35"
+indexmap = "2.7.0"
 meilisearch-auth = { path = "../meilisearch-auth" }
 meilisearch-types = { path = "../meilisearch-types" }
+memmap2 = "0.9.5"
 page_size = "0.6.0"
-raw-collections = { git = "https://github.com/meilisearch/raw-collections.git", version = "0.1.0" }
 rayon = "1.10.0"
-roaring = { version = "0.10.6", features = ["serde"] }
-serde = { version = "1.0.204", features = ["derive"] }
-serde_json = { version = "1.0.120", features = ["preserve_order"] }
+roaring = { version = "0.10.10", features = ["serde"] }
+serde = { version = "1.0.217", features = ["derive"] }
+serde_json = { version = "1.0.138", features = ["preserve_order"] }
 synchronoise = "1.0.1"
-tempfile = "3.10.1"
-thiserror = "1.0.61"
-memmap2 = "0.9.4"
-time = { version = "0.3.36", features = [
+tempfile = "3.15.0"
+thiserror = "2.0.9"
+time = { version = "0.3.37", features = [
     "serde-well-known",
     "formatting",
     "parsing",
     "macros",
 ] }
-tracing = "0.1.40"
-ureq = "2.10.0"
-uuid = { version = "1.10.0", features = ["serde", "v4"] }
-bumpalo = "3.16.0"
+tracing = "0.1.41"
+ureq = "2.12.1"
+uuid = { version = "1.11.0", features = ["serde", "v4"] }
 
 [dev-dependencies]
-arroy = "0.5.0"
 big_s = "1.0.2"
-crossbeam = "0.8.4"
-insta = { version = "1.39.0", features = ["json", "redactions"] }
+crossbeam-channel = "0.5.14"
+# fixed version due to format breakages in v1.40
+insta = { version = "=1.39.0", features = ["json", "redactions"] }
 maplit = "1.0.2"
 meili-snap = { path = "../meili-snap" }

crates/index-scheduler/src/dump.rs

@@ -0,0 +1,304 @@
+use std::collections::HashMap;
+use std::io;
+
+use dump::{KindDump, TaskDump, UpdateFile};
+use meilisearch_types::batches::{Batch, BatchId};
+use meilisearch_types::heed::RwTxn;
+use meilisearch_types::milli;
+use meilisearch_types::tasks::{Kind, KindWithContent, Status, Task};
+use roaring::RoaringBitmap;
+use uuid::Uuid;
+
+use crate::{utils, Error, IndexScheduler, Result};
+
+pub struct Dump<'a> {
+    index_scheduler: &'a IndexScheduler,
+    wtxn: RwTxn<'a>,
+
+    batch_to_task_mapping: HashMap<BatchId, RoaringBitmap>,
+
+    indexes: HashMap<String, RoaringBitmap>,
+    statuses: HashMap<Status, RoaringBitmap>,
+    kinds: HashMap<Kind, RoaringBitmap>,
+
+    batch_indexes: HashMap<String, RoaringBitmap>,
+    batch_statuses: HashMap<Status, RoaringBitmap>,
+    batch_kinds: HashMap<Kind, RoaringBitmap>,
+}
+
+impl<'a> Dump<'a> {
+    pub(crate) fn new(index_scheduler: &'a mut IndexScheduler) -> Result<Self> {
+        // While loading a dump no one should be able to access the scheduler thus I can block everything.
+        let wtxn = index_scheduler.env.write_txn()?;
+
+        Ok(Dump {
+            index_scheduler,
+            wtxn,
+            batch_to_task_mapping: HashMap::new(),
+            indexes: HashMap::new(),
+            statuses: HashMap::new(),
+            kinds: HashMap::new(),
+            batch_indexes: HashMap::new(),
+            batch_statuses: HashMap::new(),
+            batch_kinds: HashMap::new(),
+        })
+    }
+
+    /// Register a new batch coming from a dump in the scheduler.
+    /// By taking a mutable ref we're pretty sure no one will ever import a dump while actix is running.
+    pub fn register_dumped_batch(&mut self, batch: Batch) -> Result<()> {
+        self.index_scheduler.queue.batches.all_batches.put(&mut self.wtxn, &batch.uid, &batch)?;
+        if let Some(enqueued_at) = batch.enqueued_at {
+            utils::insert_task_datetime(
+                &mut self.wtxn,
+                self.index_scheduler.queue.batches.enqueued_at,
+                enqueued_at.earliest,
+                batch.uid,
+            )?;
+            utils::insert_task_datetime(
+                &mut self.wtxn,
+                self.index_scheduler.queue.batches.enqueued_at,
+                enqueued_at.oldest,
+                batch.uid,
+            )?;
+        }
+        utils::insert_task_datetime(
+            &mut self.wtxn,
+            self.index_scheduler.queue.batches.started_at,
+            batch.started_at,
+            batch.uid,
+        )?;
+        if let Some(finished_at) = batch.finished_at {
+            utils::insert_task_datetime(
+                &mut self.wtxn,
+                self.index_scheduler.queue.batches.finished_at,
+                finished_at,
+                batch.uid,
+            )?;
+        }
+
+        for index in batch.stats.index_uids.keys() {
+            match self.batch_indexes.get_mut(index) {
+                Some(bitmap) => {
+                    bitmap.insert(batch.uid);
+                }
+                None => {
+                    let mut bitmap = RoaringBitmap::new();
+                    bitmap.insert(batch.uid);
+                    self.batch_indexes.insert(index.to_string(), bitmap);
+                }
+            };
+        }
+
+        for status in batch.stats.status.keys() {
+            self.batch_statuses.entry(*status).or_default().insert(batch.uid);
+        }
+        for kind in batch.stats.types.keys() {
+            self.batch_kinds.entry(*kind).or_default().insert(batch.uid);
+        }
+
+        Ok(())
+    }
+
+    /// Register a new task coming from a dump in the scheduler.
+    /// By taking a mutable ref we're pretty sure no one will ever import a dump while actix is running.
+    pub fn register_dumped_task(
+        &mut self,
+        task: TaskDump,
+        content_file: Option<Box<UpdateFile>>,
+    ) -> Result<Task> {
+        let task_has_no_docs = matches!(task.kind, KindDump::DocumentImport { documents_count, .. } if documents_count == 0);
+
+        let content_uuid = match content_file {
+            Some(content_file) if task.status == Status::Enqueued => {
+                let (uuid, file) = self.index_scheduler.queue.create_update_file(false)?;
+                let mut writer = io::BufWriter::new(file);
+                for doc in content_file {
+                    let doc = doc?;
+                    serde_json::to_writer(&mut writer, &doc).map_err(|e| {
+                        Error::from_milli(milli::InternalError::SerdeJson(e).into(), None)
+                    })?;
+                }
+                let file = writer.into_inner().map_err(|e| e.into_error())?;
+                file.persist()?;
+
+                Some(uuid)
+            }
+            // If the task isn't `Enqueued` then just generate a recognisable `Uuid`
+            // in case we try to open it later.
+            _ if task.status != Status::Enqueued => Some(Uuid::nil()),
+            None if task.status == Status::Enqueued && task_has_no_docs => {
+                let (uuid, file) = self.index_scheduler.queue.create_update_file(false)?;
+                file.persist()?;
+
+                Some(uuid)
+            }
+            _ => None,
+        };
+
+        let task = Task {
+            uid: task.uid,
+            batch_uid: task.batch_uid,
+            enqueued_at: task.enqueued_at,
+            started_at: task.started_at,
+            finished_at: task.finished_at,
+            error: task.error,
+            canceled_by: task.canceled_by,
+            details: task.details,
+            status: task.status,
+            kind: match task.kind {
+                KindDump::DocumentImport {
+                    primary_key,
+                    method,
+                    documents_count,
+                    allow_index_creation,
+                } => KindWithContent::DocumentAdditionOrUpdate {
+                    index_uid: task.index_uid.ok_or(Error::CorruptedDump)?,
+                    primary_key,
+                    method,
+                    content_file: content_uuid.ok_or(Error::CorruptedDump)?,
+                    documents_count,
+                    allow_index_creation,
+                },
+                KindDump::DocumentDeletion { documents_ids } => KindWithContent::DocumentDeletion {
+                    documents_ids,
+                    index_uid: task.index_uid.ok_or(Error::CorruptedDump)?,
+                },
+                KindDump::DocumentDeletionByFilter { filter } => {
+                    KindWithContent::DocumentDeletionByFilter {
+                        filter_expr: filter,
+                        index_uid: task.index_uid.ok_or(Error::CorruptedDump)?,
+                    }
+                }
+                KindDump::DocumentEdition { filter, context, function } => {
+                    KindWithContent::DocumentEdition {
+                        index_uid: task.index_uid.ok_or(Error::CorruptedDump)?,
+                        filter_expr: filter,
+                        context,
+                        function,
+                    }
+                }
+                KindDump::DocumentClear => KindWithContent::DocumentClear {
+                    index_uid: task.index_uid.ok_or(Error::CorruptedDump)?,
+                },
+                KindDump::Settings { settings, is_deletion, allow_index_creation } => {
+                    KindWithContent::SettingsUpdate {
+                        index_uid: task.index_uid.ok_or(Error::CorruptedDump)?,
+                        new_settings: settings,
+                        is_deletion,
+                        allow_index_creation,
+                    }
+                }
+                KindDump::IndexDeletion => KindWithContent::IndexDeletion {
+                    index_uid: task.index_uid.ok_or(Error::CorruptedDump)?,
+                },
+                KindDump::IndexCreation { primary_key } => KindWithContent::IndexCreation {
+                    index_uid: task.index_uid.ok_or(Error::CorruptedDump)?,
+                    primary_key,
+                },
+                KindDump::IndexUpdate { primary_key } => KindWithContent::IndexUpdate {
+                    index_uid: task.index_uid.ok_or(Error::CorruptedDump)?,
+                    primary_key,
+                },
+                KindDump::IndexSwap { swaps } => KindWithContent::IndexSwap { swaps },
+                KindDump::TaskCancelation { query, tasks } => {
+                    KindWithContent::TaskCancelation { query, tasks }
+                }
+                KindDump::TasksDeletion { query, tasks } => {
+                    KindWithContent::TaskDeletion { query, tasks }
+                }
+                KindDump::DumpCreation { keys, instance_uid } => {
+                    KindWithContent::DumpCreation { keys, instance_uid }
+                }
+                KindDump::SnapshotCreation => KindWithContent::SnapshotCreation,
+                KindDump::UpgradeDatabase { from } => KindWithContent::UpgradeDatabase { from },
+            },
+        };
+
+        self.index_scheduler.queue.tasks.all_tasks.put(&mut self.wtxn, &task.uid, &task)?;
+        if let Some(batch_id) = task.batch_uid {
+            self.batch_to_task_mapping.entry(batch_id).or_default().insert(task.uid);
+        }
+
+        for index in task.indexes() {
+            match self.indexes.get_mut(index) {
+                Some(bitmap) => {
+                    bitmap.insert(task.uid);
+                }
+                None => {
+                    let mut bitmap = RoaringBitmap::new();
+                    bitmap.insert(task.uid);
+                    self.indexes.insert(index.to_string(), bitmap);
+                }
+            };
+        }
+
+        utils::insert_task_datetime(
+            &mut self.wtxn,
+            self.index_scheduler.queue.tasks.enqueued_at,
+            task.enqueued_at,
+            task.uid,
+        )?;
+
+        // we can't override the started_at & finished_at, so we must only set it if the tasks is finished and won't change
+        if matches!(task.status, Status::Succeeded | Status::Failed | Status::Canceled) {
+            if let Some(started_at) = task.started_at {
+                utils::insert_task_datetime(
+                    &mut self.wtxn,
+                    self.index_scheduler.queue.tasks.started_at,
+                    started_at,
+                    task.uid,
+                )?;
+            }
+            if let Some(finished_at) = task.finished_at {
+                utils::insert_task_datetime(
+                    &mut self.wtxn,
+                    self.index_scheduler.queue.tasks.finished_at,
+                    finished_at,
+                    task.uid,
+                )?;
+            }
+        }
+
+        self.statuses.entry(task.status).or_default().insert(task.uid);
+        self.kinds.entry(task.kind.as_kind()).or_default().insert(task.uid);
+
+        Ok(task)
+    }
+
+    /// Commit all the changes and exit the importing dump state
+    pub fn finish(mut self) -> Result<()> {
+        for (batch_id, task_ids) in self.batch_to_task_mapping {
+            self.index_scheduler.queue.batch_to_tasks_mapping.put(
+                &mut self.wtxn,
+                &batch_id,
+                &task_ids,
+            )?;
+        }
+
+        for (index, bitmap) in self.indexes {
+            self.index_scheduler.queue.tasks.index_tasks.put(&mut self.wtxn, &index, &bitmap)?;
+        }
+        for (status, bitmap) in self.statuses {
+            self.index_scheduler.queue.tasks.put_status(&mut self.wtxn, status, &bitmap)?;
+        }
+        for (kind, bitmap) in self.kinds {
+            self.index_scheduler.queue.tasks.put_kind(&mut self.wtxn, kind, &bitmap)?;
+        }
+
+        for (index, bitmap) in self.batch_indexes {
+            self.index_scheduler.queue.batches.index_tasks.put(&mut self.wtxn, &index, &bitmap)?;
+        }
+        for (status, bitmap) in self.batch_statuses {
+            self.index_scheduler.queue.batches.put_status(&mut self.wtxn, status, &bitmap)?;
+        }
+        for (kind, bitmap) in self.batch_kinds {
+            self.index_scheduler.queue.batches.put_kind(&mut self.wtxn, kind, &bitmap)?;
+        }
+
+        self.wtxn.commit()?;
+        self.index_scheduler.scheduler.wake_up.signal();
+
+        Ok(())
+    }
+}

crates/index-scheduler/src/error.rs

@@ -104,11 +104,13 @@ pub enum Error {
     )]
     InvalidTaskCanceledBy { canceled_by: String },
     #[error(
-        "{index_uid} is not a valid index uid. Index uid can be an integer or a string containing only alphanumeric characters, hyphens (-) and underscores (_), and can not be more than 512 bytes."
+        "{index_uid} is not a valid index uid. Index uid can be an integer or a string containing only alphanumeric characters, hyphens (-) and underscores (_), and can not be more than 400 bytes."
     )]
     InvalidIndexUid { index_uid: String },
     #[error("Task `{0}` not found.")]
     TaskNotFound(TaskId),
+    #[error("Task `{0}` does not contain any documents. Only `documentAdditionOrUpdate` tasks with the statuses `enqueued` or `processing` contain documents")]
+    TaskFileNotFound(TaskId),
     #[error("Batch `{0}` not found.")]
     BatchNotFound(BatchId),
     #[error("Query parameters to filter the tasks to delete are missing. Available query parameters are: `uids`, `indexUids`, `statuses`, `types`, `canceledBy`, `beforeEnqueuedAt`, `afterEnqueuedAt`, `beforeStartedAt`, `afterStartedAt`, `beforeFinishedAt`, `afterFinishedAt`.")]
@@ -122,10 +124,13 @@ pub enum Error {
     Dump(#[from] dump::Error),
     #[error(transparent)]
     Heed(#[from] heed::Error),
-    #[error(transparent)]
-    Milli(#[from] milli::Error),
-    #[error("An unexpected crash occurred when processing the task.")]
-    ProcessBatchPanicked,
+    #[error("{}", match .index_uid {
+        Some(uid) if !uid.is_empty() => format!("Index `{}`: {error}", uid),
+        _ => format!("{error}")
+    })]
+    Milli { error: milli::Error, index_uid: Option<String> },
+    #[error("An unexpected crash occurred when processing the task: {0}")]
+    ProcessBatchPanicked(String),
     #[error(transparent)]
     FileStore(#[from] file_store::Error),
     #[error(transparent)]
@@ -144,7 +149,9 @@ pub enum Error {
     #[error("Corrupted task queue.")]
     CorruptedTaskQueue,
     #[error(transparent)]
-    TaskDatabaseUpdate(Box<Self>),
+    DatabaseUpgrade(Box<Self>),
+    #[error(transparent)]
+    UnrecoverableError(Box<Self>),
     #[error(transparent)]
     HeedTransaction(heed::Error),
 
@@ -184,14 +191,15 @@ impl Error {
             | Error::InvalidTaskCanceledBy { .. }
             | Error::InvalidIndexUid { .. }
             | Error::TaskNotFound(_)
+            | Error::TaskFileNotFound(_)
             | Error::BatchNotFound(_)
             | Error::TaskDeletionWithEmptyQuery
             | Error::TaskCancelationWithEmptyQuery
             | Error::AbortedTask
             | Error::Dump(_)
             | Error::Heed(_)
-            | Error::Milli(_)
-            | Error::ProcessBatchPanicked
+            | Error::Milli { .. }
+            | Error::ProcessBatchPanicked(_)
             | Error::FileStore(_)
             | Error::IoError(_)
             | Error::Persist(_)
@@ -199,7 +207,8 @@ impl Error {
             | Error::Anyhow(_) => true,
             Error::CreateBatch(_)
             | Error::CorruptedTaskQueue
-            | Error::TaskDatabaseUpdate(_)
+            | Error::DatabaseUpgrade(_)
+            | Error::UnrecoverableError(_)
             | Error::HeedTransaction(_) => false,
             #[cfg(test)]
             Error::PlannedFailure => false,
@@ -209,6 +218,20 @@ impl Error {
     pub fn with_custom_error_code(self, code: Code) -> Self {
         Self::WithCustomErrorCode(code, Box::new(self))
     }
+
+    pub fn from_milli(err: milli::Error, index_uid: Option<String>) -> Self {
+        match err {
+            milli::Error::UserError(milli::UserError::InvalidFilter(_)) => {
+                Self::Milli { error: err, index_uid }
+                    .with_custom_error_code(Code::InvalidDocumentFilter)
+            }
+            milli::Error::UserError(milli::UserError::InvalidFilterExpression { .. }) => {
+                Self::Milli { error: err, index_uid }
+                    .with_custom_error_code(Code::InvalidDocumentFilter)
+            }
+            _ => Self::Milli { error: err, index_uid },
+        }
+    }
 }
 
 impl ErrorCode for Error {
@@ -230,14 +253,15 @@ impl ErrorCode for Error {
             Error::InvalidTaskCanceledBy { .. } => Code::InvalidTaskCanceledBy,
             Error::InvalidIndexUid { .. } => Code::InvalidIndexUid,
             Error::TaskNotFound(_) => Code::TaskNotFound,
+            Error::TaskFileNotFound(_) => Code::TaskFileNotFound,
             Error::BatchNotFound(_) => Code::BatchNotFound,
             Error::TaskDeletionWithEmptyQuery => Code::MissingTaskFilters,
             Error::TaskCancelationWithEmptyQuery => Code::MissingTaskFilters,
             // TODO: not sure of the Code to use
             Error::NoSpaceLeftInTaskQueue => Code::NoSpaceLeftOnDevice,
             Error::Dump(e) => e.error_code(),
-            Error::Milli(e) => e.error_code(),
-            Error::ProcessBatchPanicked => Code::Internal,
+            Error::Milli { error, .. } => error.error_code(),
+            Error::ProcessBatchPanicked(_) => Code::Internal,
             Error::Heed(e) => e.error_code(),
             Error::HeedTransaction(e) => e.error_code(),
             Error::FileStore(e) => e.error_code(),
@@ -249,7 +273,8 @@ impl ErrorCode for Error {
             Error::Anyhow(_) => Code::Internal,
             Error::CorruptedTaskQueue => Code::Internal,
             Error::CorruptedDump => Code::Internal,
-            Error::TaskDatabaseUpdate(_) => Code::Internal,
+            Error::DatabaseUpgrade(_) => Code::Internal,
+            Error::UnrecoverableError(_) => Code::Internal,
             Error::CreateBatch(_) => Code::Internal,
 
             // This one should never be seen by the end user

crates/index-scheduler/src/features.rs

@@ -1,18 +1,29 @@
 use std::sync::{Arc, RwLock};
 
-use meilisearch_types::features::{InstanceTogglableFeatures, RuntimeTogglableFeatures};
+use meilisearch_types::features::{InstanceTogglableFeatures, Network, RuntimeTogglableFeatures};
 use meilisearch_types::heed::types::{SerdeJson, Str};
-use meilisearch_types::heed::{Database, Env, RwTxn};
+use meilisearch_types::heed::{Database, Env, RwTxn, WithoutTls};
 
 use crate::error::FeatureNotEnabledError;
 use crate::Result;
 
-const EXPERIMENTAL_FEATURES: &str = "experimental-features";
+/// The number of database used by features
+const NUMBER_OF_DATABASES: u32 = 1;
+/// Database const names for the `FeatureData`.
+mod db_name {
+    pub const EXPERIMENTAL_FEATURES: &str = "experimental-features";
+}
+
+mod db_keys {
+    pub const EXPERIMENTAL_FEATURES: &str = "experimental-features";
+    pub const NETWORK: &str = "network";
+}
 
 #[derive(Clone)]
 pub(crate) struct FeatureData {
     persisted: Database<Str, SerdeJson<RuntimeTogglableFeatures>>,
     runtime: Arc<RwLock<RuntimeTogglableFeatures>>,
+    network: Arc<RwLock<Network>>,
 }
 
 #[derive(Debug, Clone, Copy)]
@@ -56,19 +67,6 @@ impl RoFeatures {
         }
     }
 
-    pub fn check_vector(&self, disabled_action: &'static str) -> Result<()> {
-        if self.runtime.vector_store {
-            Ok(())
-        } else {
-            Err(FeatureNotEnabledError {
-                disabled_action,
-                feature: "vector store",
-                issue_link: "https://github.com/meilisearch/product/discussions/677",
-            }
-            .into())
-        }
-    }
-
     pub fn check_edit_documents_by_function(&self, disabled_action: &'static str) -> Result<()> {
         if self.runtime.edit_documents_by_function {
             Ok(())
@@ -94,17 +92,62 @@ impl RoFeatures {
             .into())
         }
     }
+
+    pub fn check_network(&self, disabled_action: &'static str) -> Result<()> {
+        if self.runtime.network {
+            Ok(())
+        } else {
+            Err(FeatureNotEnabledError {
+                disabled_action,
+                feature: "network",
+                issue_link: "https://github.com/orgs/meilisearch/discussions/805",
+            }
+            .into())
+        }
+    }
+
+    pub fn check_get_task_documents_route(&self) -> Result<()> {
+        if self.runtime.get_task_documents_route {
+            Ok(())
+        } else {
+            Err(FeatureNotEnabledError {
+                disabled_action: "Getting the documents of an enqueued task",
+                feature: "get task documents route",
+                issue_link: "https://github.com/orgs/meilisearch/discussions/808",
+            }
+            .into())
+        }
+    }
+
+    pub fn check_composite_embedders(&self, disabled_action: &'static str) -> Result<()> {
+        if self.runtime.composite_embedders {
+            Ok(())
+        } else {
+            Err(FeatureNotEnabledError {
+                disabled_action,
+                feature: "composite embedders",
+                issue_link: "https://github.com/orgs/meilisearch/discussions/816",
+            }
+            .into())
+        }
+    }
 }
 
 impl FeatureData {
-    pub fn new(env: &Env, instance_features: InstanceTogglableFeatures) -> Result<Self> {
-        let mut wtxn = env.write_txn()?;
-        let runtime_features_db = env.create_database(&mut wtxn, Some(EXPERIMENTAL_FEATURES))?;
-        wtxn.commit()?;
+    pub(crate) const fn nb_db() -> u32 {
+        NUMBER_OF_DATABASES
+    }
+
+    pub fn new(
+        env: &Env<WithoutTls>,
+        wtxn: &mut RwTxn,
+        instance_features: InstanceTogglableFeatures,
+    ) -> Result<Self> {
+        let runtime_features_db =
+            env.create_database(wtxn, Some(db_name::EXPERIMENTAL_FEATURES))?;
 
-        let txn = env.read_txn()?;
         let persisted_features: RuntimeTogglableFeatures =
-            runtime_features_db.get(&txn, EXPERIMENTAL_FEATURES)?.unwrap_or_default();
+            runtime_features_db.get(wtxn, db_keys::EXPERIMENTAL_FEATURES)?.unwrap_or_default();
         let InstanceTogglableFeatures { metrics, logs_route, contains_filter } = instance_features;
         let runtime = Arc::new(RwLock::new(RuntimeTogglableFeatures {
             metrics: metrics || persisted_features.metrics,
@@ -113,7 +156,14 @@ impl FeatureData {
             ..persisted_features
         }));
 
-        Ok(Self { persisted: runtime_features_db, runtime })
+        let network_db = runtime_features_db.remap_data_type::<SerdeJson<Network>>();
+        let network: Network = network_db.get(wtxn, db_keys::NETWORK)?.unwrap_or_default();
+
+        Ok(Self {
+            persisted: runtime_features_db,
+            runtime,
+            network: Arc::new(RwLock::new(network)),
+        })
     }
 
     pub fn put_runtime_features(
@@ -121,7 +171,7 @@ impl FeatureData {
         mut wtxn: RwTxn,
         features: RuntimeTogglableFeatures,
     ) -> Result<()> {
-        self.persisted.put(&mut wtxn, EXPERIMENTAL_FEATURES, &features)?;
+        self.persisted.put(&mut wtxn, db_keys::EXPERIMENTAL_FEATURES, &features)?;
         wtxn.commit()?;
 
         // safe to unwrap, the lock will only fail if:
@@ -142,4 +192,21 @@ impl FeatureData {
     pub fn features(&self) -> RoFeatures {
         RoFeatures::new(self)
     }
+
+    pub fn put_network(&self, mut wtxn: RwTxn, new_network: Network) -> Result<()> {
+        self.persisted.remap_data_type::<SerdeJson<Network>>().put(
+            &mut wtxn,
+            db_keys::NETWORK,
+            &new_network,
+        )?;
+        wtxn.commit()?;
+
+        let mut network = self.network.write().unwrap();
+        *network = new_network;
+        Ok(())
+    }
+
+    pub fn network(&self) -> Network {
+        Network::clone(&*self.network.read().unwrap())
+    }
 }

crates/index-scheduler/src/index_mapper/index_map.rs

@@ -1,16 +1,17 @@
 use std::collections::BTreeMap;
+use std::env::VarError;
 use std::path::Path;
+use std::str::FromStr;
 use std::time::Duration;
 
 use meilisearch_types::heed::{EnvClosingEvent, EnvFlags, EnvOpenOptions};
-use meilisearch_types::milli::Index;
+use meilisearch_types::milli::{Index, Result};
 use time::OffsetDateTime;
 use uuid::Uuid;
 
 use super::IndexStatus::{self, Available, BeingDeleted, Closing, Missing};
+use crate::clamp_to_page_size;
 use crate::lru::{InsertionOutcome, LruMap};
-use crate::{clamp_to_page_size, Result};
-
 /// Keep an internally consistent view of the open indexes in memory.
 ///
 /// This view is made of an LRU cache that will evict the least frequently used indexes when new indexes are opened.
@@ -103,7 +104,7 @@ impl ReopenableIndex {
                 return Ok(());
             }
             map.unavailable.remove(&self.uuid);
-            map.create(&self.uuid, path, None, self.enable_mdb_writemap, self.map_size)?;
+            map.create(&self.uuid, path, None, self.enable_mdb_writemap, self.map_size, false)?;
         }
         Ok(())
     }
@@ -172,11 +173,12 @@ impl IndexMap {
         date: Option<(OffsetDateTime, OffsetDateTime)>,
         enable_mdb_writemap: bool,
         map_size: usize,
+        creation: bool,
     ) -> Result<Index> {
         if !matches!(self.get_unavailable(uuid), Missing) {
             panic!("Attempt to open an index that was unavailable");
         }
-        let index = create_or_open_index(path, date, enable_mdb_writemap, map_size)?;
+        let index = create_or_open_index(path, date, enable_mdb_writemap, map_size, creation)?;
         match self.available.insert(*uuid, index.clone()) {
             InsertionOutcome::InsertedNew => (),
             InsertionOutcome::Evicted(evicted_uuid, evicted_index) => {
@@ -300,18 +302,31 @@ fn create_or_open_index(
     date: Option<(OffsetDateTime, OffsetDateTime)>,
     enable_mdb_writemap: bool,
     map_size: usize,
+    creation: bool,
 ) -> Result<Index> {
-    let mut options = EnvOpenOptions::new();
+    let options = EnvOpenOptions::new();
+    let mut options = options.read_txn_without_tls();
     options.map_size(clamp_to_page_size(map_size));
-    options.max_readers(1024);
+
+    // You can find more details about this experimental
+    // environment variable on the following GitHub discussion:
+    // <https://github.com/orgs/meilisearch/discussions/806>
+    let max_readers = match std::env::var("MEILI_EXPERIMENTAL_INDEX_MAX_READERS") {
+        Ok(value) => u32::from_str(&value).unwrap(),
+        Err(VarError::NotPresent) => 1024,
+        Err(VarError::NotUnicode(value)) => panic!(
+            "Invalid unicode for the `MEILI_EXPERIMENTAL_INDEX_MAX_READERS` env var: {value:?}"
+        ),
+    };
+    options.max_readers(max_readers);
     if enable_mdb_writemap {
         unsafe { options.flags(EnvFlags::WRITE_MAP) };
     }
 
     if let Some((created, updated)) = date {
-        Ok(Index::new_with_creation_dates(options, path, created, updated)?)
+        Ok(Index::new_with_creation_dates(options, path, created, updated, creation)?)
     } else {
-        Ok(Index::new(options, path)?)
+        Ok(Index::new(options, path, creation)?)
     }
 }
 
@@ -319,17 +334,17 @@ fn create_or_open_index(
 #[cfg(test)]
 mod tests {
 
-    use meilisearch_types::heed::Env;
+    use meilisearch_types::heed::{Env, WithoutTls};
     use meilisearch_types::Index;
     use uuid::Uuid;
 
     use super::super::IndexMapper;
-    use crate::tests::IndexSchedulerHandle;
+    use crate::test_utils::IndexSchedulerHandle;
     use crate::utils::clamp_to_page_size;
     use crate::IndexScheduler;
 
     impl IndexMapper {
-        fn test() -> (Self, Env, IndexSchedulerHandle) {
+        fn test() -> (Self, Env<WithoutTls>, IndexSchedulerHandle) {
             let (index_scheduler, handle) = IndexScheduler::test(true, vec![]);
             (index_scheduler.index_mapper, index_scheduler.env, handle)
         }

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

@@ -4,7 +4,9 @@ use std::time::Duration;
 use std::{fs, thread};
 
 use meilisearch_types::heed::types::{SerdeJson, Str};
-use meilisearch_types::heed::{Database, Env, RoTxn, RwTxn};
+use meilisearch_types::heed::{Database, Env, RoTxn, RwTxn, WithoutTls};
+use meilisearch_types::milli;
+use meilisearch_types::milli::database_stats::DatabaseStats;
 use meilisearch_types::milli::update::IndexerConfig;
 use meilisearch_types::milli::{FieldDistribution, Index};
 use serde::{Deserialize, Serialize};
@@ -15,12 +17,17 @@ use uuid::Uuid;
 use self::index_map::IndexMap;
 use self::IndexStatus::{Available, BeingDeleted, Closing, Missing};
 use crate::uuid_codec::UuidCodec;
-use crate::{Error, Result};
+use crate::{Error, IndexBudget, IndexSchedulerOptions, Result};
 
 mod index_map;
 
-const INDEX_MAPPING: &str = "index-mapping";
-const INDEX_STATS: &str = "index-stats";
+/// The number of database used by index mapper
+const NUMBER_OF_DATABASES: u32 = 2;
+/// Database const names for the `IndexMapper`.
+mod db_name {
+    pub const INDEX_MAPPING: &str = "index-mapping";
+    pub const INDEX_STATS: &str = "index-stats";
+}
 
 /// Structure managing meilisearch's indexes.
 ///
@@ -92,19 +99,32 @@ pub enum IndexStatus {
 /// The statistics that can be computed from an `Index` object.
 #[derive(Serialize, Deserialize, Debug)]
 pub struct IndexStats {
-    /// Number of documents in the index.
-    pub number_of_documents: u64,
+    /// Stats of the documents database.
+    #[serde(default)]
+    pub documents_database_stats: DatabaseStats,
+
+    #[serde(default, skip_serializing)]
+    pub number_of_documents: Option<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.
     pub database_size: u64,
+    /// Number of embeddings in the index.
+    /// Option: retrocompatible with the stats of the pre-v1.13.0 versions of meilisearch
+    pub number_of_embeddings: Option<u64>,
+    /// Number of embedded documents in the index.
+    /// Option: retrocompatible with the stats of the pre-v1.13.0 versions of meilisearch
+    pub number_of_embedded_documents: Option<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,
+    /// The primary key of the index
+    pub primary_key: Option<String>,
     /// Association of every field name with the number of times it occurs in the documents.
     pub field_distribution: FieldDistribution,
     /// Creation date of the index.
@@ -121,11 +141,16 @@ impl IndexStats {
     /// # Parameters
     ///
     /// - rtxn: a RO transaction for the index, obtained from `Index::read_txn()`.
-    pub fn new(index: &Index, rtxn: &RoTxn) -> Result<Self> {
+    pub fn new(index: &Index, rtxn: &RoTxn) -> milli::Result<Self> {
+        let arroy_stats = index.arroy_stats(rtxn)?;
         Ok(IndexStats {
-            number_of_documents: index.number_of_documents(rtxn)?,
+            number_of_embeddings: Some(arroy_stats.number_of_embeddings),
+            number_of_embedded_documents: Some(arroy_stats.documents.len()),
+            documents_database_stats: index.documents_stats(rtxn)?.unwrap_or_default(),
+            number_of_documents: None,
             database_size: index.on_disk_size()?,
             used_database_size: index.used_size()?,
+            primary_key: index.primary_key(rtxn)?.map(|s| s.to_string()),
             field_distribution: index.field_distribution(rtxn)?,
             created_at: index.created_at(rtxn)?,
             updated_at: index.updated_at(rtxn)?,
@@ -134,29 +159,25 @@ impl IndexStats {
 }
 
 impl IndexMapper {
-    pub fn new(
-        env: &Env,
-        base_path: PathBuf,
-        index_base_map_size: usize,
-        index_growth_amount: usize,
-        index_count: usize,
-        enable_mdb_writemap: bool,
-        indexer_config: IndexerConfig,
-    ) -> Result<Self> {
-        let mut wtxn = env.write_txn()?;
-        let index_mapping = env.create_database(&mut wtxn, Some(INDEX_MAPPING))?;
-        let index_stats = env.create_database(&mut wtxn, Some(INDEX_STATS))?;
-        wtxn.commit()?;
+    pub(crate) const fn nb_db() -> u32 {
+        NUMBER_OF_DATABASES
+    }
 
+    pub fn new(
+        env: &Env<WithoutTls>,
+        wtxn: &mut RwTxn,
+        options: &IndexSchedulerOptions,
+        budget: IndexBudget,
+    ) -> Result<Self> {
         Ok(Self {
-            index_map: Arc::new(RwLock::new(IndexMap::new(index_count))),
-            index_mapping,
-            index_stats,
-            base_path,
-            index_base_map_size,
-            index_growth_amount,
-            enable_mdb_writemap,
-            indexer_config: Arc::new(indexer_config),
+            index_map: Arc::new(RwLock::new(IndexMap::new(budget.index_count))),
+            index_mapping: env.create_database(wtxn, Some(db_name::INDEX_MAPPING))?,
+            index_stats: env.create_database(wtxn, Some(db_name::INDEX_STATS))?,
+            base_path: options.indexes_path.clone(),
+            index_base_map_size: budget.map_size,
+            index_growth_amount: options.index_growth_amount,
+            enable_mdb_writemap: options.enable_mdb_writemap,
+            indexer_config: options.indexer_config.clone(),
             currently_updating_index: Default::default(),
         })
     }
@@ -183,13 +204,24 @@ impl IndexMapper {
                 // Error if the UUIDv4 somehow already exists in the map, since it should be fresh.
                 // This is very unlikely to happen in practice.
                 // TODO: it would be better to lazily create the index. But we need an Index::open function for milli.
-                let index = self.index_map.write().unwrap().create(
-                    &uuid,
-                    &index_path,
-                    date,
-                    self.enable_mdb_writemap,
-                    self.index_base_map_size,
-                )?;
+                let index = self
+                    .index_map
+                    .write()
+                    .unwrap()
+                    .create(
+                        &uuid,
+                        &index_path,
+                        date,
+                        self.enable_mdb_writemap,
+                        self.index_base_map_size,
+                        true,
+                    )
+                    .map_err(|e| Error::from_milli(e, Some(uuid.to_string())))?;
+                let index_rtxn = index.read_txn()?;
+                let stats = crate::index_mapper::IndexStats::new(&index, &index_rtxn)
+                    .map_err(|e| Error::from_milli(e, Some(name.to_string())))?;
+                self.store_stats_of(&mut wtxn, name, &stats)?;
+                drop(index_rtxn);
 
                 wtxn.commit()?;
 
@@ -357,7 +389,9 @@ impl IndexMapper {
                     };
                     let index_path = self.base_path.join(uuid.to_string());
                     // take the lock to reopen the environment.
-                    reopen.reopen(&mut self.index_map.write().unwrap(), &index_path)?;
+                    reopen
+                        .reopen(&mut self.index_map.write().unwrap(), &index_path)
+                        .map_err(|e| Error::from_milli(e, Some(uuid.to_string())))?;
                     continue;
                 }
                 BeingDeleted => return Err(Error::IndexNotFound(name.to_string())),
@@ -372,13 +406,16 @@ impl IndexMapper {
                         Missing => {
                             let index_path = self.base_path.join(uuid.to_string());
 
-                            break index_map.create(
-                                &uuid,
-                                &index_path,
-                                None,
-                                self.enable_mdb_writemap,
-                                self.index_base_map_size,
-                            )?;
+                            break index_map
+                                .create(
+                                    &uuid,
+                                    &index_path,
+                                    None,
+                                    self.enable_mdb_writemap,
+                                    self.index_base_map_size,
+                                    false,
+                                )
+                                .map_err(|e| Error::from_milli(e, Some(uuid.to_string())))?;
                         }
                         Available(index) => break index,
                         Closing(_) => {
@@ -460,6 +497,7 @@ impl IndexMapper {
                 let index = self.index(rtxn, index_uid)?;
                 let index_rtxn = index.read_txn()?;
                 IndexStats::new(&index, &index_rtxn)
+                    .map_err(|e| Error::from_milli(e, Some(uuid.to_string())))
             }
         }
     }

crates/index-scheduler/src/insta_snapshot.rs

@@ -1,15 +1,16 @@
 use std::collections::BTreeSet;
 use std::fmt::Write;
 
-use meilisearch_types::batches::Batch;
+use meilisearch_types::batches::{Batch, BatchEnqueuedAt, BatchStats};
 use meilisearch_types::heed::types::{SerdeBincode, SerdeJson, Str};
 use meilisearch_types::heed::{Database, RoTxn};
 use meilisearch_types::milli::{CboRoaringBitmapCodec, RoaringBitmapCodec, BEU32};
-use meilisearch_types::tasks::{Details, Task};
+use meilisearch_types::tasks::{Details, Kind, Status, Task};
+use meilisearch_types::versioning;
 use roaring::RoaringBitmap;
 
 use crate::index_mapper::IndexMapper;
-use crate::{IndexScheduler, Kind, Status, BEI128};
+use crate::{IndexScheduler, BEI128};
 
 pub fn snapshot_index_scheduler(scheduler: &IndexScheduler) -> String {
     // Since we'll snapshot the index right afterward, we don't need to ensure it's internally consistent for every run.
@@ -18,41 +19,15 @@ pub fn snapshot_index_scheduler(scheduler: &IndexScheduler) -> String {
     scheduler.assert_internally_consistent();
 
     let IndexScheduler {
-        autobatching_enabled,
         cleanup_enabled: _,
-        must_stop_processing: _,
         processing_tasks,
-        file_store,
         env,
-        all_tasks,
-        all_batches,
-        batch_to_tasks_mapping,
-        // task reverse index
-        status,
-        kind,
-        index_tasks,
-        canceled_by,
-        enqueued_at,
-        started_at,
-        finished_at,
-
-        // batch reverse index
-        batch_status,
-        batch_kind,
-        batch_index_tasks,
-        batch_enqueued_at,
-        batch_started_at,
-        batch_finished_at,
+        version,
+        queue,
+        scheduler,
 
         index_mapper,
         features: _,
-        max_number_of_tasks: _,
-        max_number_of_batched_tasks: _,
-        wake_up: _,
-        dumps_path: _,
-        snapshots_path: _,
-        auth_path: _,
-        version_file_path: _,
         webhook_url: _,
         webhook_authorization_header: _,
         test_breakpoint_sdr: _,
@@ -65,8 +40,18 @@ pub fn snapshot_index_scheduler(scheduler: &IndexScheduler) -> String {
 
     let mut snap = String::new();
 
+    let indx_sched_version = version.get_version(&rtxn).unwrap();
+    let latest_version = (
+        versioning::VERSION_MAJOR.parse().unwrap(),
+        versioning::VERSION_MINOR.parse().unwrap(),
+        versioning::VERSION_PATCH.parse().unwrap(),
+    );
+    if indx_sched_version != Some(latest_version) {
+        snap.push_str(&format!("index scheduler running on version {indx_sched_version:?}\n"));
+    }
+
     let processing = processing_tasks.read().unwrap().clone();
-    snap.push_str(&format!("### Autobatching Enabled = {autobatching_enabled}\n"));
+    snap.push_str(&format!("### Autobatching Enabled = {}\n", scheduler.autobatching_enabled));
     snap.push_str(&format!(
         "### Processing batch {:?}:\n",
         processing.batch.as_ref().map(|batch| batch.uid)
@@ -79,19 +64,19 @@ pub fn snapshot_index_scheduler(scheduler: &IndexScheduler) -> String {
     snap.push_str("\n----------------------------------------------------------------------\n");
 
     snap.push_str("### All Tasks:\n");
-    snap.push_str(&snapshot_all_tasks(&rtxn, *all_tasks));
+    snap.push_str(&snapshot_all_tasks(&rtxn, queue.tasks.all_tasks));
     snap.push_str("----------------------------------------------------------------------\n");
 
     snap.push_str("### Status:\n");
-    snap.push_str(&snapshot_status(&rtxn, *status));
+    snap.push_str(&snapshot_status(&rtxn, queue.tasks.status));
     snap.push_str("----------------------------------------------------------------------\n");
 
     snap.push_str("### Kind:\n");
-    snap.push_str(&snapshot_kind(&rtxn, *kind));
+    snap.push_str(&snapshot_kind(&rtxn, queue.tasks.kind));
     snap.push_str("----------------------------------------------------------------------\n");
 
     snap.push_str("### Index Tasks:\n");
-    snap.push_str(&snapshot_index_tasks(&rtxn, *index_tasks));
+    snap.push_str(&snapshot_index_tasks(&rtxn, queue.tasks.index_tasks));
     snap.push_str("----------------------------------------------------------------------\n");
 
     snap.push_str("### Index Mapper:\n");
@@ -99,55 +84,55 @@ pub fn snapshot_index_scheduler(scheduler: &IndexScheduler) -> String {
     snap.push_str("\n----------------------------------------------------------------------\n");
 
     snap.push_str("### Canceled By:\n");
-    snap.push_str(&snapshot_canceled_by(&rtxn, *canceled_by));
+    snap.push_str(&snapshot_canceled_by(&rtxn, queue.tasks.canceled_by));
     snap.push_str("\n----------------------------------------------------------------------\n");
 
     snap.push_str("### Enqueued At:\n");
-    snap.push_str(&snapshot_date_db(&rtxn, *enqueued_at));
+    snap.push_str(&snapshot_date_db(&rtxn, queue.tasks.enqueued_at));
     snap.push_str("----------------------------------------------------------------------\n");
 
     snap.push_str("### Started At:\n");
-    snap.push_str(&snapshot_date_db(&rtxn, *started_at));
+    snap.push_str(&snapshot_date_db(&rtxn, queue.tasks.started_at));
     snap.push_str("----------------------------------------------------------------------\n");
 
     snap.push_str("### Finished At:\n");
-    snap.push_str(&snapshot_date_db(&rtxn, *finished_at));
+    snap.push_str(&snapshot_date_db(&rtxn, queue.tasks.finished_at));
     snap.push_str("----------------------------------------------------------------------\n");
 
     snap.push_str("### All Batches:\n");
-    snap.push_str(&snapshot_all_batches(&rtxn, *all_batches));
+    snap.push_str(&snapshot_all_batches(&rtxn, queue.batches.all_batches));
     snap.push_str("----------------------------------------------------------------------\n");
 
     snap.push_str("### Batch to tasks mapping:\n");
-    snap.push_str(&snapshot_batches_to_tasks_mappings(&rtxn, *batch_to_tasks_mapping));
+    snap.push_str(&snapshot_batches_to_tasks_mappings(&rtxn, queue.batch_to_tasks_mapping));
     snap.push_str("----------------------------------------------------------------------\n");
 
     snap.push_str("### Batches Status:\n");
-    snap.push_str(&snapshot_status(&rtxn, *batch_status));
+    snap.push_str(&snapshot_status(&rtxn, queue.batches.status));
     snap.push_str("----------------------------------------------------------------------\n");
 
     snap.push_str("### Batches Kind:\n");
-    snap.push_str(&snapshot_kind(&rtxn, *batch_kind));
+    snap.push_str(&snapshot_kind(&rtxn, queue.batches.kind));
     snap.push_str("----------------------------------------------------------------------\n");
 
     snap.push_str("### Batches Index Tasks:\n");
-    snap.push_str(&snapshot_index_tasks(&rtxn, *batch_index_tasks));
+    snap.push_str(&snapshot_index_tasks(&rtxn, queue.batches.index_tasks));
     snap.push_str("----------------------------------------------------------------------\n");
 
     snap.push_str("### Batches Enqueued At:\n");
-    snap.push_str(&snapshot_date_db(&rtxn, *batch_enqueued_at));
+    snap.push_str(&snapshot_date_db(&rtxn, queue.batches.enqueued_at));
     snap.push_str("----------------------------------------------------------------------\n");
 
     snap.push_str("### Batches Started At:\n");
-    snap.push_str(&snapshot_date_db(&rtxn, *batch_started_at));
+    snap.push_str(&snapshot_date_db(&rtxn, queue.batches.started_at));
     snap.push_str("----------------------------------------------------------------------\n");
 
     snap.push_str("### Batches Finished At:\n");
-    snap.push_str(&snapshot_date_db(&rtxn, *batch_finished_at));
+    snap.push_str(&snapshot_date_db(&rtxn, queue.batches.finished_at));
     snap.push_str("----------------------------------------------------------------------\n");
 
     snap.push_str("### File Store:\n");
-    snap.push_str(&snapshot_file_store(file_store));
+    snap.push_str(&snapshot_file_store(&queue.file_store));
     snap.push_str("\n----------------------------------------------------------------------\n");
 
     snap
@@ -306,6 +291,9 @@ fn snapshot_details(d: &Details) -> String {
         Details::IndexSwap { swaps } => {
             format!("{{ swaps: {swaps:?} }}")
         }
+        Details::UpgradeDatabase { from, to } => {
+            format!("{{ from: {from:?}, to: {to:?} }}")
+        }
     }
 }
 
@@ -353,14 +341,24 @@ pub fn snapshot_canceled_by(rtxn: &RoTxn, db: Database<BEU32, RoaringBitmapCodec
 
 pub fn snapshot_batch(batch: &Batch) -> String {
     let mut snap = String::new();
-    let Batch { uid, details, stats, started_at, finished_at } = batch;
+    let Batch { uid, details, stats, started_at, finished_at, progress: _, enqueued_at } = batch;
+    let stats = BatchStats {
+        progress_trace: Default::default(),
+        internal_database_sizes: Default::default(),
+        write_channel_congestion: None,
+        ..stats.clone()
+    };
     if let Some(finished_at) = finished_at {
         assert!(finished_at > started_at);
     }
+    let BatchEnqueuedAt { earliest, oldest } = enqueued_at.unwrap();
+    assert!(*started_at > earliest);
+    assert!(earliest >= oldest);
+
     snap.push('{');
     snap.push_str(&format!("uid: {uid}, "));
     snap.push_str(&format!("details: {}, ", serde_json::to_string(details).unwrap()));
-    snap.push_str(&format!("stats: {}, ", serde_json::to_string(stats).unwrap()));
+    snap.push_str(&format!("stats: {}, ", serde_json::to_string(&stats).unwrap()));
     snap.push('}');
     snap
 }
@@ -373,7 +371,8 @@ pub fn snapshot_index_mapper(rtxn: &RoTxn, mapper: &IndexMapper) -> String {
         let stats = mapper.stats_of(rtxn, &name).unwrap();
         s.push_str(&format!(
             "{name}: {{ number_of_documents: {}, field_distribution: {:?} }}\n",
-            stats.number_of_documents, stats.field_distribution
+            stats.documents_database_stats.number_of_entries(),
+            stats.field_distribution
         ));
     }
 

crates/index-scheduler/src/processing.rs

@@ -0,0 +1,296 @@
+use std::sync::Arc;
+
+use meilisearch_types::milli::progress::{AtomicSubStep, NamedStep, Progress, ProgressView};
+use meilisearch_types::milli::{make_atomic_progress, make_enum_progress};
+use roaring::RoaringBitmap;
+
+use crate::utils::ProcessingBatch;
+
+#[derive(Clone, Default)]
+pub struct ProcessingTasks {
+    pub batch: Option<Arc<ProcessingBatch>>,
+    /// The list of tasks ids that are currently running.
+    pub processing: Arc<RoaringBitmap>,
+    /// The progress on processing tasks
+    pub progress: Option<Progress>,
+}
+
+impl ProcessingTasks {
+    /// Creates an empty `ProcessingAt` struct.
+    pub fn new() -> ProcessingTasks {
+        ProcessingTasks::default()
+    }
+
+    pub fn get_progress_view(&self) -> Option<ProgressView> {
+        Some(self.progress.as_ref()?.as_progress_view())
+    }
+
+    /// Stores the currently processing tasks, and the date time at which it started.
+    pub fn start_processing(
+        &mut self,
+        processing_batch: ProcessingBatch,
+        processing: RoaringBitmap,
+    ) -> Progress {
+        self.batch = Some(Arc::new(processing_batch));
+        self.processing = Arc::new(processing);
+        let progress = Progress::default();
+        progress.update_progress(BatchProgress::ProcessingTasks);
+        self.progress = Some(progress.clone());
+
+        progress
+    }
+
+    /// Set the processing tasks to an empty list
+    pub fn stop_processing(&mut self) -> Self {
+        self.progress = None;
+
+        Self {
+            batch: std::mem::take(&mut self.batch),
+            processing: std::mem::take(&mut self.processing),
+            progress: None,
+        }
+    }
+
+    /// Returns `true` if there, at least, is one task that is currently processing that we must stop.
+    pub fn must_cancel_processing_tasks(&self, canceled_tasks: &RoaringBitmap) -> bool {
+        !self.processing.is_disjoint(canceled_tasks)
+    }
+}
+
+make_enum_progress! {
+    pub enum BatchProgress {
+        ProcessingTasks,
+        WritingTasksToDisk,
+    }
+}
+
+make_enum_progress! {
+    pub enum FinalizingIndexStep {
+        Committing,
+        ComputingStats,
+    }
+}
+
+make_enum_progress! {
+    pub enum TaskCancelationProgress {
+        RetrievingTasks,
+        UpdatingTasks,
+    }
+}
+
+make_enum_progress! {
+    pub enum TaskDeletionProgress {
+        DeletingTasksDateTime,
+        DeletingTasksMetadata,
+        DeletingTasks,
+        DeletingBatches,
+    }
+}
+
+make_enum_progress! {
+    pub enum SnapshotCreationProgress {
+        StartTheSnapshotCreation,
+        SnapshotTheIndexScheduler,
+        SnapshotTheUpdateFiles,
+        SnapshotTheIndexes,
+        SnapshotTheApiKeys,
+        CreateTheTarball,
+    }
+}
+
+make_enum_progress! {
+    pub enum DumpCreationProgress {
+        StartTheDumpCreation,
+        DumpTheApiKeys,
+        DumpTheTasks,
+        DumpTheBatches,
+        DumpTheIndexes,
+        DumpTheExperimentalFeatures,
+        CompressTheDump,
+    }
+}
+
+make_enum_progress! {
+    pub enum CreateIndexProgress {
+        CreatingTheIndex,
+    }
+}
+
+make_enum_progress! {
+    pub enum UpdateIndexProgress {
+        UpdatingTheIndex,
+    }
+}
+
+make_enum_progress! {
+    pub enum DeleteIndexProgress {
+        DeletingTheIndex,
+    }
+}
+
+make_enum_progress! {
+    pub enum SwappingTheIndexes {
+        EnsuringCorrectnessOfTheSwap,
+        SwappingTheIndexes,
+    }
+}
+
+make_enum_progress! {
+    pub enum InnerSwappingTwoIndexes {
+        RetrieveTheTasks,
+        UpdateTheTasks,
+        UpdateTheIndexesMetadata,
+    }
+}
+
+make_enum_progress! {
+    pub enum DocumentOperationProgress {
+        RetrievingConfig,
+        ComputingDocumentChanges,
+        Indexing,
+    }
+}
+
+make_enum_progress! {
+    pub enum DocumentEditionProgress {
+        RetrievingConfig,
+        ComputingDocumentChanges,
+        Indexing,
+    }
+}
+
+make_enum_progress! {
+    pub enum DocumentDeletionProgress {
+        RetrievingConfig,
+        DeleteDocuments,
+        Indexing,
+    }
+}
+
+make_enum_progress! {
+    pub enum SettingsProgress {
+        RetrievingAndMergingTheSettings,
+        ApplyTheSettings,
+    }
+}
+
+make_atomic_progress!(Task alias AtomicTaskStep => "task" );
+make_atomic_progress!(Document alias AtomicDocumentStep => "document" );
+make_atomic_progress!(Batch alias AtomicBatchStep => "batch" );
+make_atomic_progress!(UpdateFile alias AtomicUpdateFileStep => "update file" );
+
+#[cfg(test)]
+mod test {
+    use std::sync::atomic::Ordering;
+
+    use meili_snap::{json_string, snapshot};
+
+    use super::*;
+
+    #[test]
+    fn one_level() {
+        let mut processing = ProcessingTasks::new();
+        processing.start_processing(ProcessingBatch::new(0), RoaringBitmap::new());
+        snapshot!(json_string!(processing.get_progress_view()), @r#"
+        {
+          "steps": [
+            {
+              "currentStep": "processing tasks",
+              "finished": 0,
+              "total": 2
+            }
+          ],
+          "percentage": 0.0
+        }
+        "#);
+        processing.progress.as_ref().unwrap().update_progress(BatchProgress::WritingTasksToDisk);
+        snapshot!(json_string!(processing.get_progress_view()), @r#"
+        {
+          "steps": [
+            {
+              "currentStep": "writing tasks to disk",
+              "finished": 1,
+              "total": 2
+            }
+          ],
+          "percentage": 50.0
+        }
+        "#);
+    }
+
+    #[test]
+    fn task_progress() {
+        let mut processing = ProcessingTasks::new();
+        processing.start_processing(ProcessingBatch::new(0), RoaringBitmap::new());
+        let (atomic, tasks) = AtomicTaskStep::new(10);
+        processing.progress.as_ref().unwrap().update_progress(tasks);
+        snapshot!(json_string!(processing.get_progress_view()), @r#"
+        {
+          "steps": [
+            {
+              "currentStep": "processing tasks",
+              "finished": 0,
+              "total": 2
+            },
+            {
+              "currentStep": "task",
+              "finished": 0,
+              "total": 10
+            }
+          ],
+          "percentage": 0.0
+        }
+        "#);
+        atomic.fetch_add(6, Ordering::Relaxed);
+        snapshot!(json_string!(processing.get_progress_view()), @r#"
+        {
+          "steps": [
+            {
+              "currentStep": "processing tasks",
+              "finished": 0,
+              "total": 2
+            },
+            {
+              "currentStep": "task",
+              "finished": 6,
+              "total": 10
+            }
+          ],
+          "percentage": 30.000002
+        }
+        "#);
+        processing.progress.as_ref().unwrap().update_progress(BatchProgress::WritingTasksToDisk);
+        snapshot!(json_string!(processing.get_progress_view()), @r#"
+        {
+          "steps": [
+            {
+              "currentStep": "writing tasks to disk",
+              "finished": 1,
+              "total": 2
+            }
+          ],
+          "percentage": 50.0
+        }
+        "#);
+        let (atomic, tasks) = AtomicTaskStep::new(5);
+        processing.progress.as_ref().unwrap().update_progress(tasks);
+        atomic.fetch_add(4, Ordering::Relaxed);
+        snapshot!(json_string!(processing.get_progress_view()), @r#"
+        {
+          "steps": [
+            {
+              "currentStep": "writing tasks to disk",
+              "finished": 1,
+              "total": 2
+            },
+            {
+              "currentStep": "task",
+              "finished": 4,
+              "total": 5
+            }
+          ],
+          "percentage": 90.0
+        }
+        "#);
+    }
+}

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

@@ -0,0 +1,603 @@
+use std::collections::HashSet;
+use std::ops::{Bound, RangeBounds};
+
+use meilisearch_types::batches::{Batch, BatchId};
+use meilisearch_types::heed::types::{DecodeIgnore, SerdeBincode, SerdeJson, Str};
+use meilisearch_types::heed::{Database, Env, RoTxn, RwTxn, WithoutTls};
+use meilisearch_types::milli::{CboRoaringBitmapCodec, RoaringBitmapCodec, BEU32};
+use meilisearch_types::tasks::{Kind, Status};
+use roaring::{MultiOps, RoaringBitmap};
+use time::OffsetDateTime;
+
+use super::{Query, Queue};
+use crate::processing::ProcessingTasks;
+use crate::utils::{
+    insert_task_datetime, keep_ids_within_datetimes, map_bound,
+    remove_n_tasks_datetime_earlier_than, remove_task_datetime, ProcessingBatch,
+};
+use crate::{Error, Result, BEI128};
+
+/// The number of database used by the batch queue
+const NUMBER_OF_DATABASES: u32 = 7;
+/// Database const names for the `IndexScheduler`.
+mod db_name {
+    pub const ALL_BATCHES: &str = "all-batches";
+
+    pub const BATCH_STATUS: &str = "batch-status";
+    pub const BATCH_KIND: &str = "batch-kind";
+    pub const BATCH_INDEX_TASKS: &str = "batch-index-tasks";
+    pub const BATCH_ENQUEUED_AT: &str = "batch-enqueued-at";
+    pub const BATCH_STARTED_AT: &str = "batch-started-at";
+    pub const BATCH_FINISHED_AT: &str = "batch-finished-at";
+}
+
+pub struct BatchQueue {
+    /// Contains all the batches accessible by their Id.
+    pub(crate) all_batches: Database<BEU32, SerdeJson<Batch>>,
+
+    /// All the batches containing a task matching the selected status.
+    pub(crate) status: Database<SerdeBincode<Status>, RoaringBitmapCodec>,
+    /// All the batches ids grouped by the kind of their task.
+    pub(crate) kind: Database<SerdeBincode<Kind>, RoaringBitmapCodec>,
+    /// Store the batches associated to an index.
+    pub(crate) index_tasks: Database<Str, RoaringBitmapCodec>,
+    /// Store the batches containing tasks which were enqueued at a specific date
+    pub(crate) enqueued_at: Database<BEI128, CboRoaringBitmapCodec>,
+    /// Store the batches containing finished tasks started at a specific date
+    pub(crate) started_at: Database<BEI128, CboRoaringBitmapCodec>,
+    /// Store the batches containing tasks finished at a specific date
+    pub(crate) finished_at: Database<BEI128, CboRoaringBitmapCodec>,
+}
+
+impl BatchQueue {
+    pub(crate) fn private_clone(&self) -> BatchQueue {
+        BatchQueue {
+            all_batches: self.all_batches,
+            status: self.status,
+            kind: self.kind,
+            index_tasks: self.index_tasks,
+            enqueued_at: self.enqueued_at,
+            started_at: self.started_at,
+            finished_at: self.finished_at,
+        }
+    }
+
+    pub(crate) const fn nb_db() -> u32 {
+        NUMBER_OF_DATABASES
+    }
+
+    pub(super) fn new(env: &Env<WithoutTls>, wtxn: &mut RwTxn) -> Result<Self> {
+        Ok(Self {
+            all_batches: env.create_database(wtxn, Some(db_name::ALL_BATCHES))?,
+            status: env.create_database(wtxn, Some(db_name::BATCH_STATUS))?,
+            kind: env.create_database(wtxn, Some(db_name::BATCH_KIND))?,
+            index_tasks: env.create_database(wtxn, Some(db_name::BATCH_INDEX_TASKS))?,
+            enqueued_at: env.create_database(wtxn, Some(db_name::BATCH_ENQUEUED_AT))?,
+            started_at: env.create_database(wtxn, Some(db_name::BATCH_STARTED_AT))?,
+            finished_at: env.create_database(wtxn, Some(db_name::BATCH_FINISHED_AT))?,
+        })
+    }
+
+    pub(crate) fn all_batch_ids(&self, rtxn: &RoTxn) -> Result<RoaringBitmap> {
+        enum_iterator::all().map(|s| self.get_status(rtxn, s)).union()
+    }
+
+    pub(crate) fn next_batch_id(&self, rtxn: &RoTxn) -> Result<BatchId> {
+        Ok(self
+            .all_batches
+            .remap_data_type::<DecodeIgnore>()
+            .last(rtxn)?
+            .map(|(k, _)| k + 1)
+            .unwrap_or_default())
+    }
+
+    pub(crate) fn get_batch(&self, rtxn: &RoTxn, batch_id: BatchId) -> Result<Option<Batch>> {
+        Ok(self.all_batches.get(rtxn, &batch_id)?)
+    }
+
+    /// Returns the whole set of batches that belongs to this index.
+    pub(crate) fn index_batches(&self, rtxn: &RoTxn, index: &str) -> Result<RoaringBitmap> {
+        Ok(self.index_tasks.get(rtxn, index)?.unwrap_or_default())
+    }
+
+    pub(crate) fn update_index(
+        &self,
+        wtxn: &mut RwTxn,
+        index: &str,
+        f: impl Fn(&mut RoaringBitmap),
+    ) -> Result<()> {
+        let mut batches = self.index_batches(wtxn, index)?;
+        f(&mut batches);
+        if batches.is_empty() {
+            self.index_tasks.delete(wtxn, index)?;
+        } else {
+            self.index_tasks.put(wtxn, index, &batches)?;
+        }
+
+        Ok(())
+    }
+
+    pub(crate) fn get_status(&self, rtxn: &RoTxn, status: Status) -> Result<RoaringBitmap> {
+        Ok(self.status.get(rtxn, &status)?.unwrap_or_default())
+    }
+
+    pub(crate) fn put_status(
+        &self,
+        wtxn: &mut RwTxn,
+        status: Status,
+        bitmap: &RoaringBitmap,
+    ) -> Result<()> {
+        Ok(self.status.put(wtxn, &status, bitmap)?)
+    }
+
+    pub(crate) fn update_status(
+        &self,
+        wtxn: &mut RwTxn,
+        status: Status,
+        f: impl Fn(&mut RoaringBitmap),
+    ) -> Result<()> {
+        let mut tasks = self.get_status(wtxn, status)?;
+        f(&mut tasks);
+        self.put_status(wtxn, status, &tasks)?;
+
+        Ok(())
+    }
+
+    pub(crate) fn get_kind(&self, rtxn: &RoTxn, kind: Kind) -> Result<RoaringBitmap> {
+        Ok(self.kind.get(rtxn, &kind)?.unwrap_or_default())
+    }
+
+    pub(crate) fn put_kind(
+        &self,
+        wtxn: &mut RwTxn,
+        kind: Kind,
+        bitmap: &RoaringBitmap,
+    ) -> Result<()> {
+        Ok(self.kind.put(wtxn, &kind, bitmap)?)
+    }
+
+    pub(crate) fn update_kind(
+        &self,
+        wtxn: &mut RwTxn,
+        kind: Kind,
+        f: impl Fn(&mut RoaringBitmap),
+    ) -> Result<()> {
+        let mut tasks = self.get_kind(wtxn, kind)?;
+        f(&mut tasks);
+        self.put_kind(wtxn, kind, &tasks)?;
+        Ok(())
+    }
+
+    pub(crate) fn write_batch(&self, wtxn: &mut RwTxn, batch: ProcessingBatch) -> Result<()> {
+        let old_batch = self.all_batches.get(wtxn, &batch.uid)?;
+
+        self.all_batches.put(
+            wtxn,
+            &batch.uid,
+            &Batch {
+                uid: batch.uid,
+                progress: None,
+                details: batch.details,
+                stats: batch.stats,
+                started_at: batch.started_at,
+                finished_at: batch.finished_at,
+                enqueued_at: batch.enqueued_at,
+            },
+        )?;
+
+        // Update the statuses
+        if let Some(ref old_batch) = old_batch {
+            for status in old_batch.stats.status.keys() {
+                self.update_status(wtxn, *status, |bitmap| {
+                    bitmap.remove(batch.uid);
+                })?;
+            }
+        }
+        for status in batch.statuses {
+            self.update_status(wtxn, status, |bitmap| {
+                bitmap.insert(batch.uid);
+            })?;
+        }
+
+        // Update the kinds / types
+        if let Some(ref old_batch) = old_batch {
+            let kinds: HashSet<_> = old_batch.stats.types.keys().cloned().collect();
+            for kind in kinds.difference(&batch.kinds) {
+                self.update_kind(wtxn, *kind, |bitmap| {
+                    bitmap.remove(batch.uid);
+                })?;
+            }
+        }
+        for kind in batch.kinds {
+            self.update_kind(wtxn, kind, |bitmap| {
+                bitmap.insert(batch.uid);
+            })?;
+        }
+
+        // Update the indexes
+        if let Some(ref old_batch) = old_batch {
+            let indexes: HashSet<_> = old_batch.stats.index_uids.keys().cloned().collect();
+            for index in indexes.difference(&batch.indexes) {
+                self.update_index(wtxn, index, |bitmap| {
+                    bitmap.remove(batch.uid);
+                })?;
+            }
+        }
+        for index in batch.indexes {
+            self.update_index(wtxn, &index, |bitmap| {
+                bitmap.insert(batch.uid);
+            })?;
+        }
+
+        // Update the enqueued_at: we cannot retrieve the previous enqueued at from the previous batch, and
+        // must instead go through the db looking for it. We cannot look at the task contained in this batch either
+        // because they may have been removed.
+        // What we know, though, is that the task date is from before the enqueued_at, and max two timestamps have been written
+        // to the DB per batches.
+        if let Some(ref old_batch) = old_batch {
+            if let Some(enqueued_at) = old_batch.enqueued_at {
+                remove_task_datetime(wtxn, self.enqueued_at, enqueued_at.earliest, old_batch.uid)?;
+                remove_task_datetime(wtxn, self.enqueued_at, enqueued_at.oldest, old_batch.uid)?;
+            } else {
+                // If we don't have the enqueued at in the batch it means the database comes from the v1.12
+                // and we still need to find the date by scrolling the database
+                remove_n_tasks_datetime_earlier_than(
+                    wtxn,
+                    self.enqueued_at,
+                    old_batch.started_at,
+                    old_batch.stats.total_nb_tasks.clamp(1, 2) as usize,
+                    old_batch.uid,
+                )?;
+            }
+        }
+        // A finished batch MUST contains at least one task and have an enqueued_at
+        let enqueued_at = batch.enqueued_at.as_ref().unwrap();
+        insert_task_datetime(wtxn, self.enqueued_at, enqueued_at.earliest, batch.uid)?;
+        insert_task_datetime(wtxn, self.enqueued_at, enqueued_at.oldest, batch.uid)?;
+
+        // Update the started at and finished at
+        if let Some(ref old_batch) = old_batch {
+            remove_task_datetime(wtxn, self.started_at, old_batch.started_at, old_batch.uid)?;
+            if let Some(finished_at) = old_batch.finished_at {
+                remove_task_datetime(wtxn, self.finished_at, finished_at, old_batch.uid)?;
+            }
+        }
+        insert_task_datetime(wtxn, self.started_at, batch.started_at, batch.uid)?;
+        insert_task_datetime(wtxn, self.finished_at, batch.finished_at.unwrap(), batch.uid)?;
+
+        Ok(())
+    }
+
+    /// Convert an iterator to a `Vec` of batches. The batches MUST exist or a
+    /// `CorruptedTaskQueue` error will be thrown.
+    pub(crate) fn get_existing_batches(
+        &self,
+        rtxn: &RoTxn,
+        tasks: impl IntoIterator<Item = BatchId>,
+        processing: &ProcessingTasks,
+    ) -> Result<Vec<Batch>> {
+        tasks
+            .into_iter()
+            .map(|batch_id| {
+                if Some(batch_id) == processing.batch.as_ref().map(|batch| batch.uid) {
+                    let mut batch = processing.batch.as_ref().unwrap().to_batch();
+                    batch.progress = processing.get_progress_view();
+                    Ok(batch)
+                } else {
+                    self.get_batch(rtxn, batch_id)
+                        .and_then(|task| task.ok_or(Error::CorruptedTaskQueue))
+                }
+            })
+            .collect::<Result<_>>()
+    }
+}
+
+impl Queue {
+    /// Return the batch ids matched by the given query from the index scheduler's point of view.
+    pub(crate) fn get_batch_ids(
+        &self,
+        rtxn: &RoTxn,
+        query: &Query,
+        processing: &ProcessingTasks,
+    ) -> Result<RoaringBitmap> {
+        let Query {
+            limit,
+            from,
+            reverse,
+            uids,
+            batch_uids,
+            statuses,
+            types,
+            index_uids,
+            canceled_by,
+            before_enqueued_at,
+            after_enqueued_at,
+            before_started_at,
+            after_started_at,
+            before_finished_at,
+            after_finished_at,
+        } = query;
+
+        let mut batches = self.batches.all_batch_ids(rtxn)?;
+        if let Some(batch_id) = processing.batch.as_ref().map(|batch| batch.uid) {
+            batches.insert(batch_id);
+        }
+
+        if let Some(from) = from {
+            let range = if reverse.unwrap_or_default() {
+                u32::MIN..*from
+            } else {
+                from.saturating_add(1)..u32::MAX
+            };
+            batches.remove_range(range);
+        }
+
+        if let Some(batch_uids) = &batch_uids {
+            let batches_uids = RoaringBitmap::from_iter(batch_uids);
+            batches &= batches_uids;
+        }
+
+        if let Some(status) = &statuses {
+            let mut status_batches = RoaringBitmap::new();
+            for status in status {
+                match status {
+                    // special case for Processing batches
+                    Status::Processing => {
+                        if let Some(batch_id) = processing.batch.as_ref().map(|batch| batch.uid) {
+                            status_batches.insert(batch_id);
+                        }
+                    }
+                    // Enqueued tasks are not stored in batches
+                    Status::Enqueued => (),
+                    status => status_batches |= &self.batches.get_status(rtxn, *status)?,
+                };
+            }
+            if !status.contains(&Status::Processing) {
+                if let Some(ref batch) = processing.batch {
+                    batches.remove(batch.uid);
+                }
+            }
+            batches &= status_batches;
+        }
+
+        if let Some(task_uids) = &uids {
+            let mut batches_by_task_uids = RoaringBitmap::new();
+            for task_uid in task_uids {
+                if let Some(task) = self.tasks.get_task(rtxn, *task_uid)? {
+                    if let Some(batch_uid) = task.batch_uid {
+                        batches_by_task_uids.insert(batch_uid);
+                    }
+                }
+            }
+            batches &= batches_by_task_uids;
+        }
+
+        // There is no database for this query, we must retrieve the task queried by the client and ensure it's valid
+        if let Some(canceled_by) = &canceled_by {
+            let mut all_canceled_batches = RoaringBitmap::new();
+            for cancel_uid in canceled_by {
+                if let Some(task) = self.tasks.get_task(rtxn, *cancel_uid)? {
+                    if task.kind.as_kind() == Kind::TaskCancelation
+                        && task.status == Status::Succeeded
+                    {
+                        if let Some(batch_uid) = task.batch_uid {
+                            all_canceled_batches.insert(batch_uid);
+                        }
+                    }
+                }
+            }
+
+            // if the canceled_by has been specified but no batch
+            // matches then we prefer matching zero than all batches.
+            if all_canceled_batches.is_empty() {
+                return Ok(RoaringBitmap::new());
+            } else {
+                batches &= all_canceled_batches;
+            }
+        }
+
+        if let Some(kind) = &types {
+            let mut kind_batches = RoaringBitmap::new();
+            for kind in kind {
+                kind_batches |= self.batches.get_kind(rtxn, *kind)?;
+                if let Some(uid) = processing
+                    .batch
+                    .as_ref()
+                    .and_then(|batch| batch.kinds.contains(kind).then_some(batch.uid))
+                {
+                    kind_batches.insert(uid);
+                }
+            }
+            batches &= &kind_batches;
+        }
+
+        if let Some(index) = &index_uids {
+            let mut index_batches = RoaringBitmap::new();
+            for index in index {
+                index_batches |= self.batches.index_batches(rtxn, index)?;
+                if let Some(uid) = processing
+                    .batch
+                    .as_ref()
+                    .and_then(|batch| batch.indexes.contains(index).then_some(batch.uid))
+                {
+                    index_batches.insert(uid);
+                }
+            }
+            batches &= &index_batches;
+        }
+
+        // For the started_at filter, we need to treat the part of the batches that are processing from the part of the
+        // batches that are not processing. The non-processing ones are filtered normally while the processing ones
+        // are entirely removed unless the in-memory startedAt variable falls within the date filter.
+        // Once we have filtered the two subsets, we put them back together and assign it back to `batches`.
+        batches = {
+            let (mut filtered_non_processing_batches, mut filtered_processing_batches) =
+                (&batches - &*processing.processing, &batches & &*processing.processing);
+
+            // special case for Processing batches
+            // A closure that clears the filtered_processing_batches if their started_at date falls outside the given bounds
+            let mut clear_filtered_processing_batches =
+                |start: Bound<OffsetDateTime>, end: Bound<OffsetDateTime>| {
+                    let start = map_bound(start, |b| b.unix_timestamp_nanos());
+                    let end = map_bound(end, |b| b.unix_timestamp_nanos());
+                    let is_within_dates = RangeBounds::contains(
+                        &(start, end),
+                        &processing
+                            .batch
+                            .as_ref()
+                            .map_or_else(OffsetDateTime::now_utc, |batch| batch.started_at)
+                            .unix_timestamp_nanos(),
+                    );
+                    if !is_within_dates {
+                        filtered_processing_batches.clear();
+                    }
+                };
+            match (after_started_at, before_started_at) {
+                (None, None) => (),
+                (None, Some(before)) => {
+                    clear_filtered_processing_batches(Bound::Unbounded, Bound::Excluded(*before))
+                }
+                (Some(after), None) => {
+                    clear_filtered_processing_batches(Bound::Excluded(*after), Bound::Unbounded)
+                }
+                (Some(after), Some(before)) => clear_filtered_processing_batches(
+                    Bound::Excluded(*after),
+                    Bound::Excluded(*before),
+                ),
+            };
+
+            keep_ids_within_datetimes(
+                rtxn,
+                &mut filtered_non_processing_batches,
+                self.batches.started_at,
+                *after_started_at,
+                *before_started_at,
+            )?;
+            filtered_non_processing_batches | filtered_processing_batches
+        };
+
+        keep_ids_within_datetimes(
+            rtxn,
+            &mut batches,
+            self.batches.enqueued_at,
+            *after_enqueued_at,
+            *before_enqueued_at,
+        )?;
+
+        keep_ids_within_datetimes(
+            rtxn,
+            &mut batches,
+            self.batches.finished_at,
+            *after_finished_at,
+            *before_finished_at,
+        )?;
+
+        if let Some(limit) = limit {
+            batches = if query.reverse.unwrap_or_default() {
+                batches.into_iter().take(*limit as usize).collect()
+            } else {
+                batches.into_iter().rev().take(*limit as usize).collect()
+            };
+        }
+
+        Ok(batches)
+    }
+
+    /// Return the batch ids matching the query along with the total number of batches
+    /// by ignoring the from and limit parameters from the user's point of view.
+    ///
+    /// There are two differences between an internal query and a query executed by
+    /// the user.
+    ///
+    /// 1. IndexSwap tasks are not publicly associated with any index, but they are associated
+    ///    with many indexes internally.
+    /// 2. The user may not have the rights to access the tasks (internally) associated with all indexes.
+    pub(crate) fn get_batch_ids_from_authorized_indexes(
+        &self,
+        rtxn: &RoTxn,
+        query: &Query,
+        filters: &meilisearch_auth::AuthFilter,
+        processing: &ProcessingTasks,
+    ) -> Result<(RoaringBitmap, u64)> {
+        // compute all batches matching the filter by ignoring the limits, to find the number of batches matching
+        // the filter.
+        // As this causes us to compute the filter twice it is slightly inefficient, but doing it this way spares
+        // us from modifying the underlying implementation, and the performance remains sufficient.
+        // Should this change, we would modify `get_batch_ids` to directly return the number of matching batches.
+        let total_batches =
+            self.get_batch_ids(rtxn, &query.clone().without_limits(), processing)?;
+        let mut batches = self.get_batch_ids(rtxn, query, processing)?;
+
+        // If the query contains a list of index uid or there is a finite list of authorized indexes,
+        // then we must exclude all the batches that only contains tasks associated to multiple indexes.
+        // This works because we don't autobatch tasks associated to multiple indexes with tasks associated
+        // to a single index. e.g: IndexSwap cannot be batched with IndexCreation.
+        if query.index_uids.is_some() || !filters.all_indexes_authorized() {
+            for kind in enum_iterator::all::<Kind>().filter(|kind| !kind.related_to_one_index()) {
+                batches -= self.tasks.get_kind(rtxn, kind)?;
+                if let Some(batch) = processing.batch.as_ref() {
+                    if batch.kinds.contains(&kind) {
+                        batches.remove(batch.uid);
+                    }
+                }
+            }
+        }
+
+        // Any batch that is internally associated with at least one authorized index
+        // must be returned.
+        if !filters.all_indexes_authorized() {
+            let mut valid_indexes = RoaringBitmap::new();
+            let mut forbidden_indexes = RoaringBitmap::new();
+
+            let all_indexes_iter = self.batches.index_tasks.iter(rtxn)?;
+            for result in all_indexes_iter {
+                let (index, index_tasks) = result?;
+                if filters.is_index_authorized(index) {
+                    valid_indexes |= index_tasks;
+                } else {
+                    forbidden_indexes |= index_tasks;
+                }
+            }
+            if let Some(batch) = processing.batch.as_ref() {
+                for index in &batch.indexes {
+                    if filters.is_index_authorized(index) {
+                        valid_indexes.insert(batch.uid);
+                    } else {
+                        forbidden_indexes.insert(batch.uid);
+                    }
+                }
+            }
+
+            // If a batch had ONE valid task then it should be returned
+            let invalid_batches = forbidden_indexes - valid_indexes;
+
+            batches -= invalid_batches;
+        }
+
+        Ok((batches, total_batches.len()))
+    }
+
+    pub(crate) fn get_batches_from_authorized_indexes(
+        &self,
+        rtxn: &RoTxn,
+        query: &Query,
+        filters: &meilisearch_auth::AuthFilter,
+        processing: &ProcessingTasks,
+    ) -> Result<(Vec<Batch>, u64)> {
+        let (batches, total) =
+            self.get_batch_ids_from_authorized_indexes(rtxn, query, filters, processing)?;
+        let batches = if query.reverse.unwrap_or_default() {
+            Box::new(batches.into_iter()) as Box<dyn Iterator<Item = u32>>
+        } else {
+            Box::new(batches.into_iter().rev()) as Box<dyn Iterator<Item = u32>>
+        };
+
+        let batches = self.batches.get_existing_batches(
+            rtxn,
+            batches.take(query.limit.unwrap_or(u32::MAX) as usize),
+            processing,
+        )?;
+
+        Ok((batches, total))
+    }
+}

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

@@ -0,0 +1,476 @@
+use meili_snap::snapshot;
+use meilisearch_auth::AuthFilter;
+use meilisearch_types::index_uid_pattern::IndexUidPattern;
+use meilisearch_types::tasks::{IndexSwap, KindWithContent, Status};
+use time::{Duration, OffsetDateTime};
+
+use crate::insta_snapshot::{snapshot_bitmap, snapshot_index_scheduler};
+use crate::test_utils::Breakpoint::*;
+use crate::test_utils::{index_creation_task, FailureLocation};
+use crate::{IndexScheduler, Query};
+
+#[test]
+fn query_batches_from_and_limit() {
+    let (index_scheduler, mut handle) = IndexScheduler::test(true, vec![]);
+
+    let kind = index_creation_task("doggo", "bone");
+    let _task = index_scheduler.register(kind, None, false).unwrap();
+    snapshot!(snapshot_index_scheduler(&index_scheduler), name: "registered_the_first_task");
+    let kind = index_creation_task("whalo", "plankton");
+    let _task = index_scheduler.register(kind, None, false).unwrap();
+    snapshot!(snapshot_index_scheduler(&index_scheduler), name: "registered_the_second_task");
+    let kind = index_creation_task("catto", "his_own_vomit");
+    let _task = index_scheduler.register(kind, None, false).unwrap();
+    snapshot!(snapshot_index_scheduler(&index_scheduler), name: "registered_the_third_task");
+
+    handle.advance_n_successful_batches(3);
+    snapshot!(snapshot_index_scheduler(&index_scheduler), name: "processed_all_tasks");
+
+    let proc = index_scheduler.processing_tasks.read().unwrap().clone();
+    let rtxn = index_scheduler.env.read_txn().unwrap();
+    let query = Query { limit: Some(0), ..Default::default() };
+    let (batches, _) = index_scheduler
+        .queue
+        .get_batch_ids_from_authorized_indexes(&rtxn, &query, &AuthFilter::default(), &proc)
+        .unwrap();
+    snapshot!(snapshot_bitmap(&batches), @"[]");
+
+    let query = Query { limit: Some(1), ..Default::default() };
+    let (batches, _) = index_scheduler
+        .queue
+        .get_batch_ids_from_authorized_indexes(&rtxn, &query, &AuthFilter::default(), &proc)
+        .unwrap();
+    snapshot!(snapshot_bitmap(&batches), @"[2,]");
+
+    let query = Query { limit: Some(2), ..Default::default() };
+    let (batches, _) = index_scheduler
+        .queue
+        .get_batch_ids_from_authorized_indexes(&rtxn, &query, &AuthFilter::default(), &proc)
+        .unwrap();
+    snapshot!(snapshot_bitmap(&batches), @"[1,2,]");
+
+    let query = Query { from: Some(1), ..Default::default() };
+    let (batches, _) = index_scheduler
+        .queue
+        .get_batch_ids_from_authorized_indexes(&rtxn, &query, &AuthFilter::default(), &proc)
+        .unwrap();
+    snapshot!(snapshot_bitmap(&batches), @"[0,1,]");
+
+    let query = Query { from: Some(2), ..Default::default() };
+    let (batches, _) = index_scheduler
+        .queue
+        .get_batch_ids_from_authorized_indexes(&rtxn, &query, &AuthFilter::default(), &proc)
+        .unwrap();
+    snapshot!(snapshot_bitmap(&batches), @"[0,1,2,]");
+
+    let query = Query { from: Some(1), limit: Some(1), ..Default::default() };
+    let (batches, _) = index_scheduler
+        .queue
+        .get_batch_ids_from_authorized_indexes(&rtxn, &query, &AuthFilter::default(), &proc)
+        .unwrap();
+    snapshot!(snapshot_bitmap(&batches), @"[1,]");
+
+    let query = Query { from: Some(1), limit: Some(2), ..Default::default() };
+    let (batches, _) = index_scheduler
+        .queue
+        .get_batch_ids_from_authorized_indexes(&rtxn, &query, &AuthFilter::default(), &proc)
+        .unwrap();
+    snapshot!(snapshot_bitmap(&batches), @"[0,1,]");
+}
+
+#[test]
+fn query_batches_simple() {
+    let start_time = OffsetDateTime::now_utc();
+
+    let (index_scheduler, mut handle) =
+        IndexScheduler::test(true, vec![(3, FailureLocation::InsideProcessBatch)]);
+
+    let kind = index_creation_task("catto", "mouse");
+    let _task = index_scheduler.register(kind, None, false).unwrap();
+    let kind = index_creation_task("doggo", "sheep");
+    let _task = index_scheduler.register(kind, None, false).unwrap();
+    let kind = index_creation_task("whalo", "fish");
+    let _task = index_scheduler.register(kind, None, false).unwrap();
+
+    snapshot!(snapshot_index_scheduler(&index_scheduler), name: "start");
+
+    handle.advance_till([Start, BatchCreated]);
+
+    let query = Query { statuses: Some(vec![Status::Processing]), ..Default::default() };
+    let (mut batches, _) = index_scheduler
+        .get_batches_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    assert_eq!(batches.len(), 1);
+    batches[0].started_at = OffsetDateTime::UNIX_EPOCH;
+    assert!(batches[0].enqueued_at.is_some());
+    batches[0].enqueued_at = None;
+    // Insta cannot snapshot our batches because the batch stats contains an enum as key: https://github.com/mitsuhiko/insta/issues/689
+    let batch = serde_json::to_string_pretty(&batches[0]).unwrap();
+    snapshot!(batch, @r#"
+    {
+      "uid": 0,
+      "details": {
+        "primaryKey": "mouse"
+      },
+      "stats": {
+        "totalNbTasks": 1,
+        "status": {
+          "processing": 1
+        },
+        "types": {
+          "indexCreation": 1
+        },
+        "indexUids": {
+          "catto": 1
+        }
+      },
+      "startedAt": "1970-01-01T00:00:00Z",
+      "finishedAt": null,
+      "enqueuedAt": null
+    }
+    "#);
+
+    let query = Query { statuses: Some(vec![Status::Enqueued]), ..Default::default() };
+    let (batches, _) = index_scheduler
+        .get_batch_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    snapshot!(snapshot_bitmap(&batches), @"[]"); // The batches don't contains any enqueued tasks
+
+    let query =
+        Query { statuses: Some(vec![Status::Enqueued, Status::Processing]), ..Default::default() };
+    let (batches, _) = index_scheduler
+        .get_batch_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    snapshot!(snapshot_bitmap(&batches), @"[0,]"); // both enqueued and processing tasks in the first tick
+
+    let query = Query {
+        statuses: Some(vec![Status::Enqueued, Status::Processing]),
+        after_started_at: Some(start_time),
+        ..Default::default()
+    };
+    let (batches, _) = index_scheduler
+        .get_batch_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    // both enqueued and processing tasks in the first tick, but limited to those with a started_at
+    // that comes after the start of the test, which should excludes the enqueued tasks
+    snapshot!(snapshot_bitmap(&batches), @"[0,]");
+
+    let query = Query {
+        statuses: Some(vec![Status::Enqueued, Status::Processing]),
+        before_started_at: Some(start_time),
+        ..Default::default()
+    };
+    let (batches, _) = index_scheduler
+        .get_batch_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    // both enqueued and processing tasks in the first tick, but limited to those with a started_at
+    // that comes before the start of the test, which should excludes all of them
+    snapshot!(snapshot_bitmap(&batches), @"[]");
+
+    let query = Query {
+        statuses: Some(vec![Status::Enqueued, Status::Processing]),
+        after_started_at: Some(start_time),
+        before_started_at: Some(start_time + Duration::minutes(1)),
+        ..Default::default()
+    };
+    let (batches, _) = index_scheduler
+        .get_batch_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    // both enqueued and processing tasks in the first tick, but limited to those with a started_at
+    // that comes after the start of the test and before one minute after the start of the test,
+    // which should exclude the enqueued tasks and include the only processing task
+    snapshot!(snapshot_bitmap(&batches), @"[0,]");
+
+    handle.advance_till([
+        InsideProcessBatch,
+        InsideProcessBatch,
+        ProcessBatchSucceeded,
+        AfterProcessing,
+        Start,
+        BatchCreated,
+    ]);
+    snapshot!(snapshot_index_scheduler(&index_scheduler), name: "after-advancing-a-bit");
+
+    let second_start_time = OffsetDateTime::now_utc();
+
+    let query = Query {
+        statuses: Some(vec![Status::Succeeded, Status::Processing]),
+        after_started_at: Some(start_time),
+        before_started_at: Some(start_time + Duration::minutes(1)),
+        ..Default::default()
+    };
+    let (batches, _) = index_scheduler
+        .get_batch_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    // both succeeded and processing tasks in the first tick, but limited to those with a started_at
+    // that comes after the start of the test and before one minute after the start of the test,
+    // which should include all tasks
+    snapshot!(snapshot_bitmap(&batches), @"[0,1,]");
+
+    let query = Query {
+        statuses: Some(vec![Status::Succeeded, Status::Processing]),
+        before_started_at: Some(start_time),
+        ..Default::default()
+    };
+    let (batches, _) = index_scheduler
+        .get_batch_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    // both succeeded and processing tasks in the first tick, but limited to those with a started_at
+    // that comes before the start of the test, which should exclude all tasks
+    snapshot!(snapshot_bitmap(&batches), @"[]");
+
+    let query = Query {
+        statuses: Some(vec![Status::Enqueued, Status::Succeeded, Status::Processing]),
+        after_started_at: Some(second_start_time),
+        before_started_at: Some(second_start_time + Duration::minutes(1)),
+        ..Default::default()
+    };
+    let (batches, _) = index_scheduler
+        .get_batch_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    // both succeeded and processing tasks in the first tick, but limited to those with a started_at
+    // that comes after the start of the second part of the test and before one minute after the
+    // second start of the test, which should exclude all tasks
+    snapshot!(snapshot_bitmap(&batches), @"[]");
+
+    // now we make one more batch, the started_at field of the new tasks will be past `second_start_time`
+    handle.advance_till([
+        InsideProcessBatch,
+        InsideProcessBatch,
+        ProcessBatchSucceeded,
+        AfterProcessing,
+        Start,
+        BatchCreated,
+    ]);
+
+    let (batches, _) = index_scheduler
+        .get_batch_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    // we run the same query to verify that, and indeed find that the last task is matched
+    snapshot!(snapshot_bitmap(&batches), @"[2,]");
+
+    let query = Query {
+        statuses: Some(vec![Status::Enqueued, Status::Succeeded, Status::Processing]),
+        after_started_at: Some(second_start_time),
+        before_started_at: Some(second_start_time + Duration::minutes(1)),
+        ..Default::default()
+    };
+    let (batches, _) = index_scheduler
+        .get_batch_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    // enqueued, succeeded, or processing tasks started after the second part of the test, should
+    // again only return the last task
+    snapshot!(snapshot_bitmap(&batches), @"[2,]");
+
+    handle.advance_till([ProcessBatchFailed, AfterProcessing]);
+
+    // now the last task should have failed
+    snapshot!(snapshot_index_scheduler(&index_scheduler), name: "end");
+    let (batches, _) = index_scheduler
+        .get_batch_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    // so running the last query should return nothing
+    snapshot!(snapshot_bitmap(&batches), @"[]");
+
+    let query = Query {
+        statuses: Some(vec![Status::Failed]),
+        after_started_at: Some(second_start_time),
+        before_started_at: Some(second_start_time + Duration::minutes(1)),
+        ..Default::default()
+    };
+    let (batches, _) = index_scheduler
+        .get_batch_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    // but the same query on failed tasks should return the last task
+    snapshot!(snapshot_bitmap(&batches), @"[2,]");
+
+    let query = Query {
+        statuses: Some(vec![Status::Failed]),
+        after_started_at: Some(second_start_time),
+        before_started_at: Some(second_start_time + Duration::minutes(1)),
+        ..Default::default()
+    };
+    let (batches, _) = index_scheduler
+        .get_batch_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    // but the same query on failed tasks should return the last task
+    snapshot!(snapshot_bitmap(&batches), @"[2,]");
+
+    let query = Query {
+        statuses: Some(vec![Status::Failed]),
+        uids: Some(vec![1]),
+        after_started_at: Some(second_start_time),
+        before_started_at: Some(second_start_time + Duration::minutes(1)),
+        ..Default::default()
+    };
+    let (batches, _) = index_scheduler
+        .get_batch_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    // same query but with an invalid uid
+    snapshot!(snapshot_bitmap(&batches), @"[]");
+
+    let query = Query {
+        statuses: Some(vec![Status::Failed]),
+        uids: Some(vec![2]),
+        after_started_at: Some(second_start_time),
+        before_started_at: Some(second_start_time + Duration::minutes(1)),
+        ..Default::default()
+    };
+    let (batches, _) = index_scheduler
+        .get_batch_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    // same query but with a valid uid
+    snapshot!(snapshot_bitmap(&batches), @"[2,]");
+}
+
+#[test]
+fn query_batches_special_rules() {
+    let (index_scheduler, mut handle) =
+        IndexScheduler::test(true, vec![(3, FailureLocation::InsideProcessBatch)]);
+
+    let kind = index_creation_task("catto", "mouse");
+    let _task = index_scheduler.register(kind, None, false).unwrap();
+    let kind = index_creation_task("doggo", "sheep");
+    let _task = index_scheduler.register(kind, None, false).unwrap();
+    let kind = KindWithContent::IndexSwap {
+        swaps: vec![IndexSwap { indexes: ("catto".to_owned(), "doggo".to_owned()) }],
+    };
+    let _task = index_scheduler.register(kind, None, false).unwrap();
+    let kind = KindWithContent::IndexSwap {
+        swaps: vec![IndexSwap { indexes: ("catto".to_owned(), "whalo".to_owned()) }],
+    };
+    let _task = index_scheduler.register(kind, None, false).unwrap();
+
+    snapshot!(snapshot_index_scheduler(&index_scheduler), name: "start");
+
+    handle.advance_till([Start, BatchCreated]);
+
+    let rtxn = index_scheduler.env.read_txn().unwrap();
+    let proc = index_scheduler.processing_tasks.read().unwrap().clone();
+
+    let query = Query { index_uids: Some(vec!["catto".to_owned()]), ..Default::default() };
+    let (batches, _) = index_scheduler
+        .queue
+        .get_batch_ids_from_authorized_indexes(&rtxn, &query, &AuthFilter::default(), &proc)
+        .unwrap();
+    // only the first task associated with catto is returned, the indexSwap tasks are excluded!
+    snapshot!(snapshot_bitmap(&batches), @"[0,]");
+
+    let query = Query { index_uids: Some(vec!["catto".to_owned()]), ..Default::default() };
+    let (batches, _) = index_scheduler
+        .queue
+        .get_batch_ids_from_authorized_indexes(
+            &rtxn,
+            &query,
+            &AuthFilter::with_allowed_indexes(
+                vec![IndexUidPattern::new_unchecked("doggo")].into_iter().collect(),
+            ),
+            &proc,
+        )
+        .unwrap();
+    // we have asked for only the tasks associated with catto, but are only authorized to retrieve the tasks
+    // associated with doggo -> empty result
+    snapshot!(snapshot_bitmap(&batches), @"[]");
+
+    drop(rtxn);
+    // We're going to advance and process all the batches for the next query to actually hit the db
+    handle.advance_till([
+        InsideProcessBatch,
+        InsideProcessBatch,
+        ProcessBatchSucceeded,
+        AfterProcessing,
+    ]);
+    handle.advance_one_successful_batch();
+    handle.advance_n_failed_batches(2);
+    snapshot!(snapshot_index_scheduler(&index_scheduler), name: "after-processing-everything");
+    let rtxn = index_scheduler.env.read_txn().unwrap();
+
+    let query = Query::default();
+    let (batches, _) = index_scheduler
+        .queue
+        .get_batch_ids_from_authorized_indexes(
+            &rtxn,
+            &query,
+            &AuthFilter::with_allowed_indexes(
+                vec![IndexUidPattern::new_unchecked("doggo")].into_iter().collect(),
+            ),
+            &proc,
+        )
+        .unwrap();
+    // we asked for all the tasks, but we are only authorized to retrieve the doggo tasks
+    // -> only the index creation of doggo should be returned
+    snapshot!(snapshot_bitmap(&batches), @"[1,]");
+
+    let query = Query::default();
+    let (batches, _) = index_scheduler
+        .queue
+        .get_batch_ids_from_authorized_indexes(
+            &rtxn,
+            &query,
+            &AuthFilter::with_allowed_indexes(
+                vec![
+                    IndexUidPattern::new_unchecked("catto"),
+                    IndexUidPattern::new_unchecked("doggo"),
+                ]
+                .into_iter()
+                .collect(),
+            ),
+            &proc,
+        )
+        .unwrap();
+    // we asked for all the tasks, but we are only authorized to retrieve the doggo and catto tasks
+    // -> all tasks except the swap of catto with whalo are returned
+    snapshot!(snapshot_bitmap(&batches), @"[0,1,]");
+
+    let query = Query::default();
+    let (batches, _) = index_scheduler
+        .queue
+        .get_batch_ids_from_authorized_indexes(&rtxn, &query, &AuthFilter::default(), &proc)
+        .unwrap();
+    // we asked for all the tasks with all index authorized -> all tasks returned
+    snapshot!(snapshot_bitmap(&batches), @"[0,1,2,3,]");
+}
+
+#[test]
+fn query_batches_canceled_by() {
+    let (index_scheduler, mut handle) =
+        IndexScheduler::test(true, vec![(3, FailureLocation::InsideProcessBatch)]);
+
+    let kind = index_creation_task("catto", "mouse");
+    let _ = index_scheduler.register(kind, None, false).unwrap();
+    let kind = index_creation_task("doggo", "sheep");
+    let _ = index_scheduler.register(kind, None, false).unwrap();
+    let kind = KindWithContent::IndexSwap {
+        swaps: vec![IndexSwap { indexes: ("catto".to_owned(), "doggo".to_owned()) }],
+    };
+    let _task = index_scheduler.register(kind, None, false).unwrap();
+
+    handle.advance_n_successful_batches(1);
+    let kind = KindWithContent::TaskCancelation {
+        query: "test_query".to_string(),
+        tasks: [0, 1, 2, 3].into_iter().collect(),
+    };
+    let task_cancelation = index_scheduler.register(kind, None, false).unwrap();
+    handle.advance_n_successful_batches(1);
+
+    snapshot!(snapshot_index_scheduler(&index_scheduler), name: "start");
+
+    let query = Query { canceled_by: Some(vec![task_cancelation.uid]), ..Query::default() };
+    let (batches, _) = index_scheduler
+        .get_batch_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    // The batch zero was the index creation task, the 1 is the task cancellation
+    snapshot!(snapshot_bitmap(&batches), @"[1,]");
+
+    let query = Query { canceled_by: Some(vec![task_cancelation.uid]), ..Query::default() };
+    let (batches, _) = index_scheduler
+        .get_batch_ids_from_authorized_indexes(
+            &query,
+            &AuthFilter::with_allowed_indexes(
+                vec![IndexUidPattern::new_unchecked("doggo")].into_iter().collect(),
+            ),
+        )
+        .unwrap();
+    // Return only 1 because the user is not authorized to see task 2
+    snapshot!(snapshot_bitmap(&batches), @"[1,]");
+}

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

@@ -0,0 +1,391 @@
+mod batches;
+#[cfg(test)]
+mod batches_test;
+mod tasks;
+#[cfg(test)]
+mod tasks_test;
+#[cfg(test)]
+mod test;
+
+use std::collections::BTreeMap;
+use std::fs::File as StdFile;
+use std::time::Duration;
+
+use file_store::FileStore;
+use meilisearch_types::batches::BatchId;
+use meilisearch_types::heed::{Database, Env, RoTxn, RwTxn, WithoutTls};
+use meilisearch_types::milli::{CboRoaringBitmapCodec, BEU32};
+use meilisearch_types::tasks::{Kind, KindWithContent, Status, Task};
+use roaring::RoaringBitmap;
+use time::format_description::well_known::Rfc3339;
+use time::OffsetDateTime;
+use uuid::Uuid;
+
+pub(crate) use self::batches::BatchQueue;
+pub(crate) use self::tasks::TaskQueue;
+use crate::processing::ProcessingTasks;
+use crate::utils::{
+    check_index_swap_validity, filter_out_references_to_newer_tasks, ProcessingBatch,
+};
+use crate::{Error, IndexSchedulerOptions, Result, TaskId};
+
+/// The number of database used by queue itself
+const NUMBER_OF_DATABASES: u32 = 1;
+/// Database const names for the `IndexScheduler`.
+mod db_name {
+    pub const BATCH_TO_TASKS_MAPPING: &str = "batch-to-tasks-mapping";
+}
+
+/// Defines a subset of tasks to be retrieved from the [`IndexScheduler`].
+///
+/// An empty/default query (where each field is set to `None`) matches all tasks.
+/// Each non-null field restricts the set of tasks further.
+#[derive(Default, Debug, Clone, PartialEq, Eq)]
+pub struct Query {
+    /// The maximum number of tasks to be matched
+    pub limit: Option<u32>,
+    /// The minimum [task id](`meilisearch_types::tasks::Task::uid`) to be matched
+    pub from: Option<u32>,
+    /// The order used to return the tasks. By default the newest tasks are returned first and the boolean is `false`.
+    pub reverse: Option<bool>,
+    /// The [task ids](`meilisearch_types::tasks::Task::uid`) to be matched
+    pub uids: Option<Vec<TaskId>>,
+    /// The [batch ids](`meilisearch_types::batches::Batch::uid`) to be matched
+    pub batch_uids: Option<Vec<BatchId>>,
+    /// The allowed [statuses](`meilisearch_types::tasks::Task::status`) of the matched tasls
+    pub statuses: Option<Vec<Status>>,
+    /// The allowed [kinds](meilisearch_types::tasks::Kind) of the matched tasks.
+    ///
+    /// The kind of a task is given by:
+    /// ```
+    /// # use meilisearch_types::tasks::{Task, Kind};
+    /// # fn doc_func(task: Task) -> Kind {
+    /// task.kind.as_kind()
+    /// # }
+    /// ```
+    pub types: Option<Vec<Kind>>,
+    /// The allowed [index ids](meilisearch_types::tasks::Task::index_uid) of the matched tasks
+    pub index_uids: Option<Vec<String>>,
+    /// The [task ids](`meilisearch_types::tasks::Task::uid`) of the [`TaskCancelation`](meilisearch_types::tasks::Task::Kind::TaskCancelation) tasks
+    /// that canceled the matched tasks.
+    pub canceled_by: Option<Vec<TaskId>>,
+    /// Exclusive upper bound of the matched tasks' [`enqueued_at`](meilisearch_types::tasks::Task::enqueued_at) field.
+    pub before_enqueued_at: Option<OffsetDateTime>,
+    /// Exclusive lower bound of the matched tasks' [`enqueued_at`](meilisearch_types::tasks::Task::enqueued_at) field.
+    pub after_enqueued_at: Option<OffsetDateTime>,
+    /// Exclusive upper bound of the matched tasks' [`started_at`](meilisearch_types::tasks::Task::started_at) field.
+    pub before_started_at: Option<OffsetDateTime>,
+    /// Exclusive lower bound of the matched tasks' [`started_at`](meilisearch_types::tasks::Task::started_at) field.
+    pub after_started_at: Option<OffsetDateTime>,
+    /// Exclusive upper bound of the matched tasks' [`finished_at`](meilisearch_types::tasks::Task::finished_at) field.
+    pub before_finished_at: Option<OffsetDateTime>,
+    /// Exclusive lower bound of the matched tasks' [`finished_at`](meilisearch_types::tasks::Task::finished_at) field.
+    pub after_finished_at: Option<OffsetDateTime>,
+}
+
+impl Query {
+    /// Return `true` if every field of the query is set to `None`, such that the query
+    /// matches all tasks.
+    pub fn is_empty(&self) -> bool {
+        matches!(
+            self,
+            Query {
+                limit: None,
+                from: None,
+                reverse: None,
+                uids: None,
+                batch_uids: None,
+                statuses: None,
+                types: None,
+                index_uids: None,
+                canceled_by: None,
+                before_enqueued_at: None,
+                after_enqueued_at: None,
+                before_started_at: None,
+                after_started_at: None,
+                before_finished_at: None,
+                after_finished_at: None,
+            }
+        )
+    }
+
+    /// Add an [index id](meilisearch_types::tasks::Task::index_uid) to the list of permitted indexes.
+    pub fn with_index(self, index_uid: String) -> Self {
+        let mut index_vec = self.index_uids.unwrap_or_default();
+        index_vec.push(index_uid);
+        Self { index_uids: Some(index_vec), ..self }
+    }
+
+    // Removes the `from` and `limit` restrictions from the query.
+    // Useful to get the total number of tasks matching a filter.
+    pub fn without_limits(self) -> Self {
+        Query { limit: None, from: None, ..self }
+    }
+}
+
+/// Structure which holds meilisearch's indexes and schedules the tasks
+/// to be performed on them.
+pub struct Queue {
+    pub(crate) tasks: tasks::TaskQueue,
+    pub(crate) batches: batches::BatchQueue,
+
+    /// Matches a batch id with the associated task ids.
+    pub(crate) batch_to_tasks_mapping: Database<BEU32, CboRoaringBitmapCodec>,
+
+    /// The list of files referenced by the tasks.
+    pub(crate) file_store: FileStore,
+
+    /// The max number of tasks allowed before the scheduler starts to delete
+    /// the finished tasks automatically.
+    pub(crate) max_number_of_tasks: usize,
+}
+
+impl Queue {
+    pub(crate) fn private_clone(&self) -> Queue {
+        Queue {
+            tasks: self.tasks.private_clone(),
+            batches: self.batches.private_clone(),
+            batch_to_tasks_mapping: self.batch_to_tasks_mapping,
+            file_store: self.file_store.clone(),
+            max_number_of_tasks: self.max_number_of_tasks,
+        }
+    }
+
+    pub(crate) const fn nb_db() -> u32 {
+        tasks::TaskQueue::nb_db() + batches::BatchQueue::nb_db() + NUMBER_OF_DATABASES
+    }
+
+    /// Create an index scheduler and start its run loop.
+    pub(crate) fn new(
+        env: &Env<WithoutTls>,
+        wtxn: &mut RwTxn,
+        options: &IndexSchedulerOptions,
+    ) -> Result<Self> {
+        // allow unreachable_code to get rids of the warning in the case of a test build.
+        Ok(Self {
+            file_store: FileStore::new(&options.update_file_path)?,
+            batch_to_tasks_mapping: env
+                .create_database(wtxn, Some(db_name::BATCH_TO_TASKS_MAPPING))?,
+            tasks: TaskQueue::new(env, wtxn)?,
+            batches: BatchQueue::new(env, wtxn)?,
+            max_number_of_tasks: options.max_number_of_tasks,
+        })
+    }
+
+    /// Returns the whole set of tasks that belongs to this batch.
+    pub(crate) fn tasks_in_batch(&self, rtxn: &RoTxn, batch_id: BatchId) -> Result<RoaringBitmap> {
+        Ok(self.batch_to_tasks_mapping.get(rtxn, &batch_id)?.unwrap_or_default())
+    }
+
+    /// Convert an iterator to a `Vec` of tasks and edit the `ProcessingBatch` to add the given tasks.
+    ///
+    /// The tasks MUST exist, or a `CorruptedTaskQueue` error will be thrown.
+    pub(crate) fn get_existing_tasks_for_processing_batch(
+        &self,
+        rtxn: &RoTxn,
+        processing_batch: &mut ProcessingBatch,
+        tasks: impl IntoIterator<Item = TaskId>,
+    ) -> Result<Vec<Task>> {
+        tasks
+            .into_iter()
+            .map(|task_id| {
+                let mut task = self
+                    .tasks
+                    .get_task(rtxn, task_id)
+                    .and_then(|task| task.ok_or(Error::CorruptedTaskQueue));
+                processing_batch.processing(&mut task);
+                task
+            })
+            .collect::<Result<_>>()
+    }
+
+    pub(crate) fn write_batch(
+        &self,
+        wtxn: &mut RwTxn,
+        batch: ProcessingBatch,
+        tasks: &RoaringBitmap,
+    ) -> Result<()> {
+        self.batch_to_tasks_mapping.put(wtxn, &batch.uid, tasks)?;
+        self.batches.write_batch(wtxn, batch)?;
+        Ok(())
+    }
+
+    pub(crate) fn delete_persisted_task_data(&self, task: &Task) -> Result<()> {
+        match task.content_uuid() {
+            Some(content_file) => self.delete_update_file(content_file),
+            None => Ok(()),
+        }
+    }
+
+    /// Open and returns the task's content File.
+    pub fn update_file(&self, uuid: Uuid) -> file_store::Result<StdFile> {
+        self.file_store.get_update(uuid)
+    }
+
+    /// Delete a file from the index scheduler.
+    ///
+    /// Counterpart to the [`create_update_file`](IndexScheduler::create_update_file) method.
+    pub fn delete_update_file(&self, uuid: Uuid) -> Result<()> {
+        Ok(self.file_store.delete(uuid)?)
+    }
+
+    /// Create a file and register it in the index scheduler.
+    ///
+    /// The returned file and uuid can be used to associate
+    /// some data to a task. The file will be kept until
+    /// the task has been fully processed.
+    pub fn create_update_file(&self, dry_run: bool) -> Result<(Uuid, file_store::File)> {
+        if dry_run {
+            Ok((Uuid::nil(), file_store::File::dry_file()?))
+        } else {
+            Ok(self.file_store.new_update()?)
+        }
+    }
+
+    #[cfg(test)]
+    pub fn create_update_file_with_uuid(&self, uuid: u128) -> Result<(Uuid, file_store::File)> {
+        Ok(self.file_store.new_update_with_uuid(uuid)?)
+    }
+
+    /// The size on disk taken by all the updates files contained in the `IndexScheduler`, in bytes.
+    pub fn compute_update_file_size(&self) -> Result<u64> {
+        Ok(self.file_store.compute_total_size()?)
+    }
+
+    pub fn register(
+        &self,
+        wtxn: &mut RwTxn,
+        kind: &KindWithContent,
+        task_id: Option<TaskId>,
+        dry_run: bool,
+    ) -> Result<Task> {
+        let next_task_id = self.tasks.next_task_id(wtxn)?;
+
+        if let Some(uid) = task_id {
+            if uid < next_task_id {
+                return Err(Error::BadTaskId { received: uid, expected: next_task_id });
+            }
+        }
+
+        let mut task = Task {
+            uid: task_id.unwrap_or(next_task_id),
+            // The batch is defined once we starts processing the task
+            batch_uid: None,
+            enqueued_at: OffsetDateTime::now_utc(),
+            started_at: None,
+            finished_at: None,
+            error: None,
+            canceled_by: None,
+            details: kind.default_details(),
+            status: Status::Enqueued,
+            kind: kind.clone(),
+        };
+        // For deletion and cancelation tasks, we want to make extra sure that they
+        // don't attempt to delete/cancel tasks that are newer than themselves.
+        filter_out_references_to_newer_tasks(&mut task);
+        // If the register task is an index swap task, verify that it is well-formed
+        // (that it does not contain duplicate indexes).
+        check_index_swap_validity(&task)?;
+
+        // At this point the task is going to be registered and no further checks will be done
+        if dry_run {
+            return Ok(task);
+        }
+
+        // Get rid of the mutability.
+        let task = task;
+        self.tasks.register(wtxn, &task)?;
+
+        Ok(task)
+    }
+
+    /// Register a task to cleanup the task queue if needed
+    pub fn cleanup_task_queue(&self, wtxn: &mut RwTxn) -> Result<()> {
+        let nb_tasks = self.tasks.all_task_ids(wtxn)?.len();
+        // if we have less than 1M tasks everything is fine
+        if nb_tasks < self.max_number_of_tasks as u64 {
+            return Ok(());
+        }
+
+        let finished = self.tasks.status.get(wtxn, &Status::Succeeded)?.unwrap_or_default()
+            | self.tasks.status.get(wtxn, &Status::Failed)?.unwrap_or_default()
+            | self.tasks.status.get(wtxn, &Status::Canceled)?.unwrap_or_default();
+
+        let to_delete = RoaringBitmap::from_iter(finished.into_iter().rev().take(100_000));
+
+        // /!\ the len must be at least 2 or else we might enter an infinite loop where we only delete
+        //     the deletion tasks we enqueued ourselves.
+        if to_delete.len() < 2 {
+            tracing::warn!("The task queue is almost full, but no task can be deleted yet.");
+            // the only thing we can do is hope that the user tasks are going to finish
+            return Ok(());
+        }
+
+        tracing::info!(
+            "The task queue is almost full. Deleting the oldest {} finished tasks.",
+            to_delete.len()
+        );
+
+        // it's safe to unwrap here because we checked the len above
+        let newest_task_id = to_delete.iter().last().unwrap();
+        let last_task_to_delete =
+            self.tasks.get_task(wtxn, newest_task_id)?.ok_or(Error::CorruptedTaskQueue)?;
+
+        // increase time by one nanosecond so that the enqueuedAt of the last task to delete is also lower than that date.
+        let delete_before = last_task_to_delete.enqueued_at + Duration::from_nanos(1);
+
+        self.register(
+            wtxn,
+            &KindWithContent::TaskDeletion {
+                query: format!(
+                    "?beforeEnqueuedAt={}&statuses=succeeded,failed,canceled",
+                    delete_before.format(&Rfc3339).map_err(|_| Error::CorruptedTaskQueue)?,
+                ),
+                tasks: to_delete,
+            },
+            None,
+            false,
+        )?;
+
+        Ok(())
+    }
+
+    pub fn get_stats(
+        &self,
+        rtxn: &RoTxn,
+        processing: &ProcessingTasks,
+    ) -> Result<BTreeMap<String, BTreeMap<String, u64>>> {
+        let mut res = BTreeMap::new();
+        let processing_tasks = processing.processing.len();
+
+        res.insert(
+            "statuses".to_string(),
+            enum_iterator::all::<Status>()
+                .map(|s| {
+                    let tasks = self.tasks.get_status(rtxn, s)?.len();
+                    match s {
+                        Status::Enqueued => Ok((s.to_string(), tasks - processing_tasks)),
+                        Status::Processing => Ok((s.to_string(), processing_tasks)),
+                        s => Ok((s.to_string(), tasks)),
+                    }
+                })
+                .collect::<Result<BTreeMap<String, u64>>>()?,
+        );
+        res.insert(
+            "types".to_string(),
+            enum_iterator::all::<Kind>()
+                .map(|s| Ok((s.to_string(), self.tasks.get_kind(rtxn, s)?.len())))
+                .collect::<Result<BTreeMap<String, u64>>>()?,
+        );
+        res.insert(
+            "indexes".to_string(),
+            self.tasks
+                .index_tasks
+                .iter(rtxn)?
+                .map(|res| Ok(res.map(|(name, bitmap)| (name.to_string(), bitmap.len()))?))
+                .collect::<Result<BTreeMap<String, u64>>>()?,
+        );
+
+        Ok(res)
+    }
+}

crates/index-scheduler/src/queue/snapshots/batches_test.rs/query_batches_canceled_by/start.snap

@@ -1,5 +1,5 @@
 ---
-source: crates/index-scheduler/src/lib.rs
+source: crates/index-scheduler/src/queue/batches_test.rs
 snapshot_kind: text
 ---
 ### Autobatching Enabled = true

crates/index-scheduler/src/queue/snapshots/batches_test.rs/query_batches_from_and_limit/processed_all_tasks.snap

@@ -1,5 +1,5 @@
 ---
-source: crates/index-scheduler/src/lib.rs
+source: crates/index-scheduler/src/queue/batches_test.rs
 snapshot_kind: text
 ---
 ### Autobatching Enabled = true

crates/index-scheduler/src/queue/snapshots/batches_test.rs/query_batches_from_and_limit/registered_the_first_task.snap

@@ -1,5 +1,5 @@
 ---
-source: crates/index-scheduler/src/lib.rs
+source: crates/index-scheduler/src/queue/batches_test.rs
 snapshot_kind: text
 ---
 ### Autobatching Enabled = true

crates/index-scheduler/src/queue/snapshots/batches_test.rs/query_batches_from_and_limit/registered_the_second_task.snap

@@ -1,5 +1,5 @@
 ---
-source: crates/index-scheduler/src/lib.rs
+source: crates/index-scheduler/src/queue/batches_test.rs
 snapshot_kind: text
 ---
 ### Autobatching Enabled = true

crates/index-scheduler/src/queue/snapshots/batches_test.rs/query_batches_from_and_limit/registered_the_third_task.snap

@@ -1,5 +1,5 @@
 ---
-source: crates/index-scheduler/src/lib.rs
+source: crates/index-scheduler/src/queue/batches_test.rs
 snapshot_kind: text
 ---
 ### Autobatching Enabled = true

crates/index-scheduler/src/queue/snapshots/batches_test.rs/query_batches_simple/after-advancing-a-bit.snap

@@ -1,11 +1,11 @@
 ---
-source: crates/index-scheduler/src/lib.rs
+source: crates/index-scheduler/src/queue/batches_test.rs
 snapshot_kind: text
 ---
 ### Autobatching Enabled = true
 ### Processing batch Some(1):
 [1,]
-{uid: 1, details: {"primaryKey":"sheep"}, stats: {"totalNbTasks":2,"status":{"enqueued":2},"types":{"indexCreation":2},"indexUids":{"doggo":2}}, }
+{uid: 1, details: {"primaryKey":"sheep"}, stats: {"totalNbTasks":1,"status":{"processing":1},"types":{"indexCreation":1},"indexUids":{"doggo":1}}, }
 ----------------------------------------------------------------------
 ### All Tasks:
 0 {uid: 0, batch_uid: 0, status: succeeded, details: { primary_key: Some("mouse") }, kind: IndexCreation { index_uid: "catto", primary_key: Some("mouse") }}

crates/index-scheduler/src/queue/snapshots/batches_test.rs/query_batches_simple/end.snap

@@ -1,5 +1,5 @@
 ---
-source: crates/index-scheduler/src/lib.rs
+source: crates/index-scheduler/src/queue/batches_test.rs
 snapshot_kind: text
 ---
 ### Autobatching Enabled = true

crates/index-scheduler/src/queue/snapshots/batches_test.rs/query_batches_simple/start.snap

@@ -1,5 +1,5 @@
 ---
-source: crates/index-scheduler/src/lib.rs
+source: crates/index-scheduler/src/queue/batches_test.rs
 snapshot_kind: text
 ---
 ### Autobatching Enabled = true

crates/index-scheduler/src/queue/snapshots/batches_test.rs/query_batches_special_rules/after-processing-everything.snap

@@ -1,5 +1,5 @@
 ---
-source: crates/index-scheduler/src/lib.rs
+source: crates/index-scheduler/src/queue/batches_test.rs
 snapshot_kind: text
 ---
 ### Autobatching Enabled = true

crates/index-scheduler/src/queue/snapshots/batches_test.rs/query_batches_special_rules/start.snap

@@ -1,5 +1,5 @@
 ---
-source: crates/index-scheduler/src/lib.rs
+source: crates/index-scheduler/src/queue/batches_test.rs
 snapshot_kind: text
 ---
 ### Autobatching Enabled = true

crates/index-scheduler/src/queue/snapshots/tasks_test.rs/query_tasks_canceled_by/start.snap

@@ -1,5 +1,5 @@
 ---
-source: crates/index-scheduler/src/lib.rs
+source: crates/index-scheduler/src/queue/tasks_test.rs
 snapshot_kind: text
 ---
 ### Autobatching Enabled = true

crates/index-scheduler/src/queue/snapshots/tasks_test.rs/query_tasks_from_and_limit/processed_all_tasks.snap

@@ -1,5 +1,5 @@
 ---
-source: crates/index-scheduler/src/lib.rs
+source: crates/index-scheduler/src/queue/tasks_test.rs
 snapshot_kind: text
 ---
 ### Autobatching Enabled = true

crates/index-scheduler/src/queue/snapshots/tasks_test.rs/query_tasks_from_and_limit/registered_the_first_task.snap

@@ -1,5 +1,5 @@
 ---
-source: crates/index-scheduler/src/lib.rs
+source: crates/index-scheduler/src/queue/tasks_test.rs
 snapshot_kind: text
 ---
 ### Autobatching Enabled = true

crates/index-scheduler/src/queue/snapshots/tasks_test.rs/query_tasks_from_and_limit/registered_the_second_task.snap

@@ -1,5 +1,5 @@
 ---
-source: crates/index-scheduler/src/lib.rs
+source: crates/index-scheduler/src/queue/tasks_test.rs
 snapshot_kind: text
 ---
 ### Autobatching Enabled = true

crates/index-scheduler/src/queue/snapshots/tasks_test.rs/query_tasks_from_and_limit/registered_the_third_task.snap

@@ -1,5 +1,5 @@
 ---
-source: crates/index-scheduler/src/lib.rs
+source: crates/index-scheduler/src/queue/tasks_test.rs
 snapshot_kind: text
 ---
 ### Autobatching Enabled = true

crates/index-scheduler/src/queue/snapshots/tasks_test.rs/query_tasks_simple/end.snap

@@ -1,5 +1,5 @@
 ---
-source: crates/index-scheduler/src/lib.rs
+source: crates/index-scheduler/src/queue/tasks_test.rs
 snapshot_kind: text
 ---
 ### Autobatching Enabled = true

crates/index-scheduler/src/queue/snapshots/tasks_test.rs/query_tasks_simple/start.snap

@@ -1,5 +1,5 @@
 ---
-source: crates/index-scheduler/src/lib.rs
+source: crates/index-scheduler/src/queue/tasks_test.rs
 snapshot_kind: text
 ---
 ### Autobatching Enabled = true

crates/index-scheduler/src/queue/snapshots/tasks_test.rs/query_tasks_special_rules/start.snap

@@ -1,5 +1,5 @@
 ---
-source: crates/index-scheduler/src/lib.rs
+source: crates/index-scheduler/src/queue/tasks_test.rs
 snapshot_kind: text
 ---
 ### Autobatching Enabled = true

crates/index-scheduler/src/queue/snapshots/test.rs/register/everything_is_successfully_registered.snap

@@ -1,5 +1,5 @@
 ---
-source: crates/index-scheduler/src/lib.rs
+source: crates/index-scheduler/src/queue/test.rs
 snapshot_kind: text
 ---
 ### Autobatching Enabled = true

crates/index-scheduler/src/queue/snapshots/test.rs/test_auto_deletion_of_tasks/after_the_second_task_deletion.snap

@@ -1,5 +1,6 @@
 ---
-source: crates/index-scheduler/src/lib.rs
+source: crates/index-scheduler/src/queue/test.rs
+snapshot_kind: text
 ---
 [
   {

crates/index-scheduler/src/queue/snapshots/test.rs/test_auto_deletion_of_tasks/everything_has_been_processed.snap

@@ -1,5 +1,6 @@
 ---
-source: crates/index-scheduler/src/lib.rs
+source: crates/index-scheduler/src/queue/test.rs
+snapshot_kind: text
 ---
 [
   {

crates/index-scheduler/src/queue/snapshots/test.rs/test_auto_deletion_of_tasks/task_deletion_have_been_enqueued.snap

@@ -1,5 +1,6 @@
 ---
-source: crates/index-scheduler/src/lib.rs
+source: crates/index-scheduler/src/queue/test.rs
+snapshot_kind: text
 ---
 [
   {

crates/index-scheduler/src/queue/snapshots/test.rs/test_auto_deletion_of_tasks/task_deletion_have_been_processed.snap

@@ -1,5 +1,6 @@
 ---
-source: crates/index-scheduler/src/lib.rs
+source: crates/index-scheduler/src/queue/test.rs
+snapshot_kind: text
 ---
 [
   {

crates/index-scheduler/src/queue/snapshots/test.rs/test_auto_deletion_of_tasks/task_queue_is_full.snap

@@ -1,5 +1,6 @@
 ---
-source: crates/index-scheduler/src/lib.rs
+source: crates/index-scheduler/src/queue/test.rs
+snapshot_kind: text
 ---
 [
   {

crates/index-scheduler/src/queue/snapshots/test.rs/test_disable_auto_deletion_of_tasks/task_deletion_have_not_been_enqueued.snap

@@ -1,5 +1,6 @@
 ---
-source: crates/index-scheduler/src/lib.rs
+source: crates/index-scheduler/src/queue/test.rs
+snapshot_kind: text
 ---
 [
   {

crates/index-scheduler/src/queue/snapshots/test.rs/test_disable_auto_deletion_of_tasks/task_queue_is_full.snap

@@ -1,5 +1,6 @@
 ---
-source: crates/index-scheduler/src/lib.rs
+source: crates/index-scheduler/src/queue/test.rs
+snapshot_kind: text
 ---
 [
   {

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

@@ -0,0 +1,541 @@
+use std::ops::{Bound, RangeBounds};
+
+use meilisearch_types::heed::types::{DecodeIgnore, SerdeBincode, SerdeJson, Str};
+use meilisearch_types::heed::{Database, Env, RoTxn, RwTxn, WithoutTls};
+use meilisearch_types::milli::{CboRoaringBitmapCodec, RoaringBitmapCodec, BEU32};
+use meilisearch_types::tasks::{Kind, Status, Task};
+use roaring::{MultiOps, RoaringBitmap};
+use time::OffsetDateTime;
+
+use super::{Query, Queue};
+use crate::processing::ProcessingTasks;
+use crate::utils::{
+    self, insert_task_datetime, keep_ids_within_datetimes, map_bound, remove_task_datetime,
+};
+use crate::{Error, Result, TaskId, BEI128};
+
+/// The number of database used by the task queue
+const NUMBER_OF_DATABASES: u32 = 8;
+/// Database const names for the `IndexScheduler`.
+mod db_name {
+    pub const ALL_TASKS: &str = "all-tasks";
+
+    pub const STATUS: &str = "status";
+    pub const KIND: &str = "kind";
+    pub const INDEX_TASKS: &str = "index-tasks";
+    pub const CANCELED_BY: &str = "canceled_by";
+    pub const ENQUEUED_AT: &str = "enqueued-at";
+    pub const STARTED_AT: &str = "started-at";
+    pub const FINISHED_AT: &str = "finished-at";
+}
+
+pub struct TaskQueue {
+    /// The main database, it contains all the tasks accessible by their Id.
+    pub(crate) all_tasks: Database<BEU32, SerdeJson<Task>>,
+
+    /// All the tasks ids grouped by their status.
+    // TODO we should not be able to serialize a `Status::Processing` in this database.
+    pub(crate) status: Database<SerdeBincode<Status>, RoaringBitmapCodec>,
+    /// All the tasks ids grouped by their kind.
+    pub(crate) kind: Database<SerdeBincode<Kind>, RoaringBitmapCodec>,
+    /// Store the tasks associated to an index.
+    pub(crate) index_tasks: Database<Str, RoaringBitmapCodec>,
+    /// Store the tasks that were canceled by a task uid
+    pub(crate) canceled_by: Database<BEU32, RoaringBitmapCodec>,
+    /// Store the task ids of tasks which were enqueued at a specific date
+    pub(crate) enqueued_at: Database<BEI128, CboRoaringBitmapCodec>,
+    /// Store the task ids of finished tasks which started being processed at a specific date
+    pub(crate) started_at: Database<BEI128, CboRoaringBitmapCodec>,
+    /// Store the task ids of tasks which finished at a specific date
+    pub(crate) finished_at: Database<BEI128, CboRoaringBitmapCodec>,
+}
+
+impl TaskQueue {
+    pub(crate) fn private_clone(&self) -> TaskQueue {
+        TaskQueue {
+            all_tasks: self.all_tasks,
+            status: self.status,
+            kind: self.kind,
+            index_tasks: self.index_tasks,
+            canceled_by: self.canceled_by,
+            enqueued_at: self.enqueued_at,
+            started_at: self.started_at,
+            finished_at: self.finished_at,
+        }
+    }
+
+    pub(crate) const fn nb_db() -> u32 {
+        NUMBER_OF_DATABASES
+    }
+
+    pub(crate) fn new(env: &Env<WithoutTls>, wtxn: &mut RwTxn) -> Result<Self> {
+        Ok(Self {
+            all_tasks: env.create_database(wtxn, Some(db_name::ALL_TASKS))?,
+            status: env.create_database(wtxn, Some(db_name::STATUS))?,
+            kind: env.create_database(wtxn, Some(db_name::KIND))?,
+            index_tasks: env.create_database(wtxn, Some(db_name::INDEX_TASKS))?,
+            canceled_by: env.create_database(wtxn, Some(db_name::CANCELED_BY))?,
+            enqueued_at: env.create_database(wtxn, Some(db_name::ENQUEUED_AT))?,
+            started_at: env.create_database(wtxn, Some(db_name::STARTED_AT))?,
+            finished_at: env.create_database(wtxn, Some(db_name::FINISHED_AT))?,
+        })
+    }
+
+    pub(crate) fn last_task_id(&self, rtxn: &RoTxn) -> Result<Option<TaskId>> {
+        Ok(self.all_tasks.remap_data_type::<DecodeIgnore>().last(rtxn)?.map(|(k, _)| k + 1))
+    }
+
+    pub(crate) fn next_task_id(&self, rtxn: &RoTxn) -> Result<TaskId> {
+        Ok(self.last_task_id(rtxn)?.unwrap_or_default())
+    }
+
+    pub(crate) fn all_task_ids(&self, rtxn: &RoTxn) -> Result<RoaringBitmap> {
+        enum_iterator::all().map(|s| self.get_status(rtxn, s)).union()
+    }
+
+    pub(crate) fn get_task(&self, rtxn: &RoTxn, task_id: TaskId) -> Result<Option<Task>> {
+        Ok(self.all_tasks.get(rtxn, &task_id)?)
+    }
+
+    pub(crate) fn update_task(&self, wtxn: &mut RwTxn, task: &Task) -> Result<()> {
+        let old_task = self.get_task(wtxn, task.uid)?.ok_or(Error::CorruptedTaskQueue)?;
+        let reprocessing = old_task.status != Status::Enqueued;
+
+        debug_assert!(old_task != *task);
+        debug_assert_eq!(old_task.uid, task.uid);
+
+        // If we're processing a task that failed it may already contains a batch_uid
+        debug_assert!(
+            reprocessing || (old_task.batch_uid.is_none() && task.batch_uid.is_some()),
+            "\n==> old: {old_task:?}\n==> new: {task:?}"
+        );
+
+        if old_task.status != task.status {
+            self.update_status(wtxn, old_task.status, |bitmap| {
+                bitmap.remove(task.uid);
+            })?;
+            self.update_status(wtxn, task.status, |bitmap| {
+                bitmap.insert(task.uid);
+            })?;
+        }
+
+        if old_task.kind.as_kind() != task.kind.as_kind() {
+            self.update_kind(wtxn, old_task.kind.as_kind(), |bitmap| {
+                bitmap.remove(task.uid);
+            })?;
+            self.update_kind(wtxn, task.kind.as_kind(), |bitmap| {
+                bitmap.insert(task.uid);
+            })?;
+        }
+
+        assert_eq!(
+            old_task.enqueued_at, task.enqueued_at,
+            "Cannot update a task's enqueued_at time"
+        );
+        if old_task.started_at != task.started_at {
+            assert!(
+                reprocessing || old_task.started_at.is_none(),
+                "Cannot update a task's started_at time"
+            );
+            if let Some(started_at) = old_task.started_at {
+                remove_task_datetime(wtxn, self.started_at, started_at, task.uid)?;
+            }
+            if let Some(started_at) = task.started_at {
+                insert_task_datetime(wtxn, self.started_at, started_at, task.uid)?;
+            }
+        }
+        if old_task.finished_at != task.finished_at {
+            assert!(
+                reprocessing || old_task.finished_at.is_none(),
+                "Cannot update a task's finished_at time"
+            );
+            if let Some(finished_at) = old_task.finished_at {
+                remove_task_datetime(wtxn, self.finished_at, finished_at, task.uid)?;
+            }
+            if let Some(finished_at) = task.finished_at {
+                insert_task_datetime(wtxn, self.finished_at, finished_at, task.uid)?;
+            }
+        }
+
+        self.all_tasks.put(wtxn, &task.uid, task)?;
+        Ok(())
+    }
+
+    /// Returns the whole set of tasks that belongs to this index.
+    pub(crate) fn index_tasks(&self, rtxn: &RoTxn, index: &str) -> Result<RoaringBitmap> {
+        Ok(self.index_tasks.get(rtxn, index)?.unwrap_or_default())
+    }
+
+    pub(crate) fn update_index(
+        &self,
+        wtxn: &mut RwTxn,
+        index: &str,
+        f: impl Fn(&mut RoaringBitmap),
+    ) -> Result<()> {
+        let mut tasks = self.index_tasks(wtxn, index)?;
+        f(&mut tasks);
+        if tasks.is_empty() {
+            self.index_tasks.delete(wtxn, index)?;
+        } else {
+            self.index_tasks.put(wtxn, index, &tasks)?;
+        }
+
+        Ok(())
+    }
+
+    pub(crate) fn get_status(&self, rtxn: &RoTxn, status: Status) -> Result<RoaringBitmap> {
+        Ok(self.status.get(rtxn, &status)?.unwrap_or_default())
+    }
+
+    pub(crate) fn put_status(
+        &self,
+        wtxn: &mut RwTxn,
+        status: Status,
+        bitmap: &RoaringBitmap,
+    ) -> Result<()> {
+        Ok(self.status.put(wtxn, &status, bitmap)?)
+    }
+
+    pub(crate) fn update_status(
+        &self,
+        wtxn: &mut RwTxn,
+        status: Status,
+        f: impl Fn(&mut RoaringBitmap),
+    ) -> Result<()> {
+        let mut tasks = self.get_status(wtxn, status)?;
+        f(&mut tasks);
+        self.put_status(wtxn, status, &tasks)?;
+
+        Ok(())
+    }
+
+    pub(crate) fn get_kind(&self, rtxn: &RoTxn, kind: Kind) -> Result<RoaringBitmap> {
+        Ok(self.kind.get(rtxn, &kind)?.unwrap_or_default())
+    }
+
+    pub(crate) fn put_kind(
+        &self,
+        wtxn: &mut RwTxn,
+        kind: Kind,
+        bitmap: &RoaringBitmap,
+    ) -> Result<()> {
+        Ok(self.kind.put(wtxn, &kind, bitmap)?)
+    }
+
+    pub(crate) fn update_kind(
+        &self,
+        wtxn: &mut RwTxn,
+        kind: Kind,
+        f: impl Fn(&mut RoaringBitmap),
+    ) -> Result<()> {
+        let mut tasks = self.get_kind(wtxn, kind)?;
+        f(&mut tasks);
+        self.put_kind(wtxn, kind, &tasks)?;
+
+        Ok(())
+    }
+
+    /// Convert an iterator to a `Vec` of tasks. The tasks MUST exist or a
+    /// `CorruptedTaskQueue` error will be thrown.
+    pub(crate) fn get_existing_tasks(
+        &self,
+        rtxn: &RoTxn,
+        tasks: impl IntoIterator<Item = TaskId>,
+    ) -> Result<Vec<Task>> {
+        tasks
+            .into_iter()
+            .map(|task_id| {
+                self.get_task(rtxn, task_id).and_then(|task| task.ok_or(Error::CorruptedTaskQueue))
+            })
+            .collect::<Result<_>>()
+    }
+
+    pub(crate) fn register(&self, wtxn: &mut RwTxn, task: &Task) -> Result<()> {
+        self.all_tasks.put(wtxn, &task.uid, task)?;
+
+        for index in task.indexes() {
+            self.update_index(wtxn, index, |bitmap| {
+                bitmap.insert(task.uid);
+            })?;
+        }
+
+        self.update_status(wtxn, Status::Enqueued, |bitmap| {
+            bitmap.insert(task.uid);
+        })?;
+
+        self.update_kind(wtxn, task.kind.as_kind(), |bitmap| {
+            bitmap.insert(task.uid);
+        })?;
+
+        utils::insert_task_datetime(wtxn, self.enqueued_at, task.enqueued_at, task.uid)?;
+
+        Ok(())
+    }
+}
+
+impl Queue {
+    /// Return the task ids matched by the given query from the index scheduler's point of view.
+    pub(crate) fn get_task_ids(
+        &self,
+        rtxn: &RoTxn,
+        query: &Query,
+        processing_tasks: &ProcessingTasks,
+    ) -> Result<RoaringBitmap> {
+        let ProcessingTasks { batch: processing_batch, processing: processing_tasks, progress: _ } =
+            processing_tasks;
+        let Query {
+            limit,
+            from,
+            reverse,
+            uids,
+            batch_uids,
+            statuses,
+            types,
+            index_uids,
+            canceled_by,
+            before_enqueued_at,
+            after_enqueued_at,
+            before_started_at,
+            after_started_at,
+            before_finished_at,
+            after_finished_at,
+        } = query;
+
+        let mut tasks = self.tasks.all_task_ids(rtxn)?;
+
+        if let Some(from) = from {
+            let range = if reverse.unwrap_or_default() {
+                u32::MIN..*from
+            } else {
+                from.saturating_add(1)..u32::MAX
+            };
+            tasks.remove_range(range);
+        }
+
+        if let Some(batch_uids) = batch_uids {
+            let mut batch_tasks = RoaringBitmap::new();
+            for batch_uid in batch_uids {
+                if processing_batch.as_ref().map_or(false, |batch| batch.uid == *batch_uid) {
+                    batch_tasks |= &**processing_tasks;
+                } else {
+                    batch_tasks |= self.tasks_in_batch(rtxn, *batch_uid)?;
+                }
+            }
+            tasks &= batch_tasks;
+        }
+
+        if let Some(status) = statuses {
+            let mut status_tasks = RoaringBitmap::new();
+            for status in status {
+                match status {
+                    // special case for Processing tasks
+                    Status::Processing => {
+                        status_tasks |= &**processing_tasks;
+                    }
+                    status => status_tasks |= &self.tasks.get_status(rtxn, *status)?,
+                };
+            }
+            if !status.contains(&Status::Processing) {
+                tasks -= &**processing_tasks;
+            }
+            tasks &= status_tasks;
+        }
+
+        if let Some(uids) = uids {
+            let uids = RoaringBitmap::from_iter(uids);
+            tasks &= &uids;
+        }
+
+        if let Some(canceled_by) = canceled_by {
+            let mut all_canceled_tasks = RoaringBitmap::new();
+            for cancel_task_uid in canceled_by {
+                if let Some(canceled_by_uid) = self.tasks.canceled_by.get(rtxn, cancel_task_uid)? {
+                    all_canceled_tasks |= canceled_by_uid;
+                }
+            }
+
+            // if the canceled_by has been specified but no task
+            // matches then we prefer matching zero than all tasks.
+            if all_canceled_tasks.is_empty() {
+                return Ok(RoaringBitmap::new());
+            } else {
+                tasks &= all_canceled_tasks;
+            }
+        }
+
+        if let Some(kind) = types {
+            let mut kind_tasks = RoaringBitmap::new();
+            for kind in kind {
+                kind_tasks |= self.tasks.get_kind(rtxn, *kind)?;
+            }
+            tasks &= &kind_tasks;
+        }
+
+        if let Some(index) = index_uids {
+            let mut index_tasks = RoaringBitmap::new();
+            for index in index {
+                index_tasks |= self.tasks.index_tasks(rtxn, index)?;
+            }
+            tasks &= &index_tasks;
+        }
+
+        // For the started_at filter, we need to treat the part of the tasks that are processing from the part of the
+        // tasks that are not processing. The non-processing ones are filtered normally while the processing ones
+        // are entirely removed unless the in-memory startedAt variable falls within the date filter.
+        // Once we have filtered the two subsets, we put them back together and assign it back to `tasks`.
+        tasks = {
+            let (mut filtered_non_processing_tasks, mut filtered_processing_tasks) =
+                (&tasks - &**processing_tasks, &tasks & &**processing_tasks);
+
+            // special case for Processing tasks
+            // A closure that clears the filtered_processing_tasks if their started_at date falls outside the given bounds
+            let mut clear_filtered_processing_tasks =
+                |start: Bound<OffsetDateTime>, end: Bound<OffsetDateTime>| {
+                    let start = map_bound(start, |b| b.unix_timestamp_nanos());
+                    let end = map_bound(end, |b| b.unix_timestamp_nanos());
+                    let is_within_dates = RangeBounds::contains(
+                        &(start, end),
+                        &processing_batch
+                            .as_ref()
+                            .map_or_else(OffsetDateTime::now_utc, |batch| batch.started_at)
+                            .unix_timestamp_nanos(),
+                    );
+                    if !is_within_dates {
+                        filtered_processing_tasks.clear();
+                    }
+                };
+            match (after_started_at, before_started_at) {
+                (None, None) => (),
+                (None, Some(before)) => {
+                    clear_filtered_processing_tasks(Bound::Unbounded, Bound::Excluded(*before))
+                }
+                (Some(after), None) => {
+                    clear_filtered_processing_tasks(Bound::Excluded(*after), Bound::Unbounded)
+                }
+                (Some(after), Some(before)) => clear_filtered_processing_tasks(
+                    Bound::Excluded(*after),
+                    Bound::Excluded(*before),
+                ),
+            };
+
+            keep_ids_within_datetimes(
+                rtxn,
+                &mut filtered_non_processing_tasks,
+                self.tasks.started_at,
+                *after_started_at,
+                *before_started_at,
+            )?;
+            filtered_non_processing_tasks | filtered_processing_tasks
+        };
+
+        keep_ids_within_datetimes(
+            rtxn,
+            &mut tasks,
+            self.tasks.enqueued_at,
+            *after_enqueued_at,
+            *before_enqueued_at,
+        )?;
+
+        keep_ids_within_datetimes(
+            rtxn,
+            &mut tasks,
+            self.tasks.finished_at,
+            *after_finished_at,
+            *before_finished_at,
+        )?;
+
+        if let Some(limit) = limit {
+            tasks = if query.reverse.unwrap_or_default() {
+                tasks.into_iter().take(*limit as usize).collect()
+            } else {
+                tasks.into_iter().rev().take(*limit as usize).collect()
+            };
+        }
+
+        Ok(tasks)
+    }
+
+    pub(crate) fn get_task_ids_from_authorized_indexes(
+        &self,
+        rtxn: &RoTxn,
+        query: &Query,
+        filters: &meilisearch_auth::AuthFilter,
+        processing_tasks: &ProcessingTasks,
+    ) -> Result<(RoaringBitmap, u64)> {
+        // compute all tasks matching the filter by ignoring the limits, to find the number of tasks matching
+        // the filter.
+        // As this causes us to compute the filter twice it is slightly inefficient, but doing it this way spares
+        // us from modifying the underlying implementation, and the performance remains sufficient.
+        // Should this change, we would modify `get_task_ids` to directly return the number of matching tasks.
+        let total_tasks =
+            self.get_task_ids(rtxn, &query.clone().without_limits(), processing_tasks)?;
+        let mut tasks = self.get_task_ids(rtxn, query, processing_tasks)?;
+
+        // If the query contains a list of index uid or there is a finite list of authorized indexes,
+        // then we must exclude all the kinds that aren't associated to one and only one index.
+        if query.index_uids.is_some() || !filters.all_indexes_authorized() {
+            for kind in enum_iterator::all::<Kind>().filter(|kind| !kind.related_to_one_index()) {
+                tasks -= self.tasks.get_kind(rtxn, kind)?;
+            }
+        }
+
+        // Any task that is internally associated with a non-authorized index
+        // must be discarded.
+        if !filters.all_indexes_authorized() {
+            let all_indexes_iter = self.tasks.index_tasks.iter(rtxn)?;
+            for result in all_indexes_iter {
+                let (index, index_tasks) = result?;
+                if !filters.is_index_authorized(index) {
+                    tasks -= index_tasks;
+                }
+            }
+        }
+
+        Ok((tasks, total_tasks.len()))
+    }
+
+    pub(crate) fn get_tasks_from_authorized_indexes(
+        &self,
+        rtxn: &RoTxn,
+        query: &Query,
+        filters: &meilisearch_auth::AuthFilter,
+        processing_tasks: &ProcessingTasks,
+    ) -> Result<(Vec<Task>, u64)> {
+        let (tasks, total) =
+            self.get_task_ids_from_authorized_indexes(rtxn, query, filters, processing_tasks)?;
+        let tasks = if query.reverse.unwrap_or_default() {
+            Box::new(tasks.into_iter()) as Box<dyn Iterator<Item = u32>>
+        } else {
+            Box::new(tasks.into_iter().rev()) as Box<dyn Iterator<Item = u32>>
+        };
+        let tasks = self
+            .tasks
+            .get_existing_tasks(rtxn, tasks.take(query.limit.unwrap_or(u32::MAX) as usize))?;
+
+        let ProcessingTasks { batch, processing, progress: _ } = processing_tasks;
+
+        let ret = tasks.into_iter();
+        if processing.is_empty() || batch.is_none() {
+            Ok((ret.collect(), total))
+        } else {
+            // Safe because we ensured there was a batch in the previous branch
+            let batch = batch.as_ref().unwrap();
+            Ok((
+                ret.map(|task| {
+                    if processing.contains(task.uid) {
+                        Task {
+                            status: Status::Processing,
+                            batch_uid: Some(batch.uid),
+                            started_at: Some(batch.started_at),
+                            ..task
+                        }
+                    } else {
+                        task
+                    }
+                })
+                .collect(),
+                total,
+            ))
+        }
+    }
+}

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

@@ -0,0 +1,441 @@
+use meili_snap::snapshot;
+use meilisearch_auth::AuthFilter;
+use meilisearch_types::index_uid_pattern::IndexUidPattern;
+use meilisearch_types::tasks::{IndexSwap, KindWithContent, Status};
+use time::{Duration, OffsetDateTime};
+
+use crate::insta_snapshot::{snapshot_bitmap, snapshot_index_scheduler};
+use crate::test_utils::Breakpoint::*;
+use crate::test_utils::{index_creation_task, FailureLocation};
+use crate::{IndexScheduler, Query};
+
+#[test]
+fn query_tasks_from_and_limit() {
+    let (index_scheduler, mut handle) = IndexScheduler::test(true, vec![]);
+
+    let kind = index_creation_task("doggo", "bone");
+    let _task = index_scheduler.register(kind, None, false).unwrap();
+    snapshot!(snapshot_index_scheduler(&index_scheduler), name: "registered_the_first_task");
+    let kind = index_creation_task("whalo", "plankton");
+    let _task = index_scheduler.register(kind, None, false).unwrap();
+    snapshot!(snapshot_index_scheduler(&index_scheduler), name: "registered_the_second_task");
+    let kind = index_creation_task("catto", "his_own_vomit");
+    let _task = index_scheduler.register(kind, None, false).unwrap();
+    snapshot!(snapshot_index_scheduler(&index_scheduler), name: "registered_the_third_task");
+
+    handle.advance_n_successful_batches(3);
+    snapshot!(snapshot_index_scheduler(&index_scheduler), name: "processed_all_tasks");
+
+    let rtxn = index_scheduler.env.read_txn().unwrap();
+    let processing = index_scheduler.processing_tasks.read().unwrap();
+    let query = Query { limit: Some(0), ..Default::default() };
+    let (tasks, _) = index_scheduler
+        .queue
+        .get_task_ids_from_authorized_indexes(&rtxn, &query, &AuthFilter::default(), &processing)
+        .unwrap();
+    snapshot!(snapshot_bitmap(&tasks), @"[]");
+
+    let query = Query { limit: Some(1), ..Default::default() };
+    let (tasks, _) = index_scheduler
+        .queue
+        .get_task_ids_from_authorized_indexes(&rtxn, &query, &AuthFilter::default(), &processing)
+        .unwrap();
+    snapshot!(snapshot_bitmap(&tasks), @"[2,]");
+
+    let query = Query { limit: Some(2), ..Default::default() };
+    let (tasks, _) = index_scheduler
+        .queue
+        .get_task_ids_from_authorized_indexes(&rtxn, &query, &AuthFilter::default(), &processing)
+        .unwrap();
+    snapshot!(snapshot_bitmap(&tasks), @"[1,2,]");
+
+    let query = Query { from: Some(1), ..Default::default() };
+    let (tasks, _) = index_scheduler
+        .queue
+        .get_task_ids_from_authorized_indexes(&rtxn, &query, &AuthFilter::default(), &processing)
+        .unwrap();
+    snapshot!(snapshot_bitmap(&tasks), @"[0,1,]");
+
+    let query = Query { from: Some(2), ..Default::default() };
+    let (tasks, _) = index_scheduler
+        .queue
+        .get_task_ids_from_authorized_indexes(&rtxn, &query, &AuthFilter::default(), &processing)
+        .unwrap();
+    snapshot!(snapshot_bitmap(&tasks), @"[0,1,2,]");
+
+    let query = Query { from: Some(1), limit: Some(1), ..Default::default() };
+    let (tasks, _) = index_scheduler
+        .queue
+        .get_task_ids_from_authorized_indexes(&rtxn, &query, &AuthFilter::default(), &processing)
+        .unwrap();
+    snapshot!(snapshot_bitmap(&tasks), @"[1,]");
+
+    let query = Query { from: Some(1), limit: Some(2), ..Default::default() };
+    let (tasks, _) = index_scheduler
+        .queue
+        .get_task_ids_from_authorized_indexes(&rtxn, &query, &AuthFilter::default(), &processing)
+        .unwrap();
+    snapshot!(snapshot_bitmap(&tasks), @"[0,1,]");
+}
+
+#[test]
+fn query_tasks_simple() {
+    let start_time = OffsetDateTime::now_utc();
+
+    let (index_scheduler, mut handle) =
+        IndexScheduler::test(true, vec![(3, FailureLocation::InsideProcessBatch)]);
+
+    let kind = index_creation_task("catto", "mouse");
+    let _task = index_scheduler.register(kind, None, false).unwrap();
+    let kind = index_creation_task("doggo", "sheep");
+    let _task = index_scheduler.register(kind, None, false).unwrap();
+    let kind = index_creation_task("whalo", "fish");
+    let _task = index_scheduler.register(kind, None, false).unwrap();
+
+    snapshot!(snapshot_index_scheduler(&index_scheduler), name: "start");
+
+    handle.advance_till([Start, BatchCreated]);
+
+    let query = Query { statuses: Some(vec![Status::Processing]), ..Default::default() };
+    let (tasks, _) = index_scheduler
+        .get_task_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    snapshot!(snapshot_bitmap(&tasks), @"[0,]"); // only the processing tasks in the first tick
+
+    let query = Query { statuses: Some(vec![Status::Enqueued]), ..Default::default() };
+    let (tasks, _) = index_scheduler
+        .get_task_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    snapshot!(snapshot_bitmap(&tasks), @"[1,2,]"); // only the enqueued tasks in the first tick
+
+    let query =
+        Query { statuses: Some(vec![Status::Enqueued, Status::Processing]), ..Default::default() };
+    let (tasks, _) = index_scheduler
+        .get_task_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    snapshot!(snapshot_bitmap(&tasks), @"[0,1,2,]"); // both enqueued and processing tasks in the first tick
+
+    let query = Query {
+        statuses: Some(vec![Status::Enqueued, Status::Processing]),
+        after_started_at: Some(start_time),
+        ..Default::default()
+    };
+    let (tasks, _) = index_scheduler
+        .get_task_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    // both enqueued and processing tasks in the first tick, but limited to those with a started_at
+    // that comes after the start of the test, which should excludes the enqueued tasks
+    snapshot!(snapshot_bitmap(&tasks), @"[0,]");
+
+    let query = Query {
+        statuses: Some(vec![Status::Enqueued, Status::Processing]),
+        before_started_at: Some(start_time),
+        ..Default::default()
+    };
+    let (tasks, _) = index_scheduler
+        .get_task_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    // both enqueued and processing tasks in the first tick, but limited to those with a started_at
+    // that comes before the start of the test, which should excludes all of them
+    snapshot!(snapshot_bitmap(&tasks), @"[]");
+
+    let query = Query {
+        statuses: Some(vec![Status::Enqueued, Status::Processing]),
+        after_started_at: Some(start_time),
+        before_started_at: Some(start_time + Duration::minutes(1)),
+        ..Default::default()
+    };
+    let (tasks, _) = index_scheduler
+        .get_task_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    // both enqueued and processing tasks in the first tick, but limited to those with a started_at
+    // that comes after the start of the test and before one minute after the start of the test,
+    // which should exclude the enqueued tasks and include the only processing task
+    snapshot!(snapshot_bitmap(&tasks), @"[0,]");
+
+    handle.advance_till([
+        InsideProcessBatch,
+        InsideProcessBatch,
+        ProcessBatchSucceeded,
+        AfterProcessing,
+        Start,
+        BatchCreated,
+    ]);
+
+    let second_start_time = OffsetDateTime::now_utc();
+
+    let query = Query {
+        statuses: Some(vec![Status::Succeeded, Status::Processing]),
+        after_started_at: Some(start_time),
+        before_started_at: Some(start_time + Duration::minutes(1)),
+        ..Default::default()
+    };
+    let (tasks, _) = index_scheduler
+        .get_task_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    // both succeeded and processing tasks in the first tick, but limited to those with a started_at
+    // that comes after the start of the test and before one minute after the start of the test,
+    // which should include all tasks
+    snapshot!(snapshot_bitmap(&tasks), @"[0,1,]");
+
+    let query = Query {
+        statuses: Some(vec![Status::Succeeded, Status::Processing]),
+        before_started_at: Some(start_time),
+        ..Default::default()
+    };
+    let (tasks, _) = index_scheduler
+        .get_task_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    // both succeeded and processing tasks in the first tick, but limited to those with a started_at
+    // that comes before the start of the test, which should exclude all tasks
+    snapshot!(snapshot_bitmap(&tasks), @"[]");
+
+    let query = Query {
+        statuses: Some(vec![Status::Enqueued, Status::Succeeded, Status::Processing]),
+        after_started_at: Some(second_start_time),
+        before_started_at: Some(second_start_time + Duration::minutes(1)),
+        ..Default::default()
+    };
+    let (tasks, _) = index_scheduler
+        .get_task_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    // both succeeded and processing tasks in the first tick, but limited to those with a started_at
+    // that comes after the start of the second part of the test and before one minute after the
+    // second start of the test, which should exclude all tasks
+    snapshot!(snapshot_bitmap(&tasks), @"[]");
+
+    // now we make one more batch, the started_at field of the new tasks will be past `second_start_time`
+    handle.advance_till([
+        InsideProcessBatch,
+        InsideProcessBatch,
+        ProcessBatchSucceeded,
+        AfterProcessing,
+        Start,
+        BatchCreated,
+    ]);
+
+    let (tasks, _) = index_scheduler
+        .get_task_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    // we run the same query to verify that, and indeed find that the last task is matched
+    snapshot!(snapshot_bitmap(&tasks), @"[2,]");
+
+    let query = Query {
+        statuses: Some(vec![Status::Enqueued, Status::Succeeded, Status::Processing]),
+        after_started_at: Some(second_start_time),
+        before_started_at: Some(second_start_time + Duration::minutes(1)),
+        ..Default::default()
+    };
+    let (tasks, _) = index_scheduler
+        .get_task_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    // enqueued, succeeded, or processing tasks started after the second part of the test, should
+    // again only return the last task
+    snapshot!(snapshot_bitmap(&tasks), @"[2,]");
+
+    handle.advance_till([ProcessBatchFailed, AfterProcessing]);
+
+    // now the last task should have failed
+    snapshot!(snapshot_index_scheduler(&index_scheduler), name: "end");
+    let (tasks, _) = index_scheduler
+        .get_task_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    // so running the last query should return nothing
+    snapshot!(snapshot_bitmap(&tasks), @"[]");
+
+    let query = Query {
+        statuses: Some(vec![Status::Failed]),
+        after_started_at: Some(second_start_time),
+        before_started_at: Some(second_start_time + Duration::minutes(1)),
+        ..Default::default()
+    };
+    let (tasks, _) = index_scheduler
+        .get_task_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    // but the same query on failed tasks should return the last task
+    snapshot!(snapshot_bitmap(&tasks), @"[2,]");
+
+    let query = Query {
+        statuses: Some(vec![Status::Failed]),
+        after_started_at: Some(second_start_time),
+        before_started_at: Some(second_start_time + Duration::minutes(1)),
+        ..Default::default()
+    };
+    let (tasks, _) = index_scheduler
+        .get_task_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    // but the same query on failed tasks should return the last task
+    snapshot!(snapshot_bitmap(&tasks), @"[2,]");
+
+    let query = Query {
+        statuses: Some(vec![Status::Failed]),
+        uids: Some(vec![1]),
+        after_started_at: Some(second_start_time),
+        before_started_at: Some(second_start_time + Duration::minutes(1)),
+        ..Default::default()
+    };
+    let (tasks, _) = index_scheduler
+        .get_task_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    // same query but with an invalid uid
+    snapshot!(snapshot_bitmap(&tasks), @"[]");
+
+    let query = Query {
+        statuses: Some(vec![Status::Failed]),
+        uids: Some(vec![2]),
+        after_started_at: Some(second_start_time),
+        before_started_at: Some(second_start_time + Duration::minutes(1)),
+        ..Default::default()
+    };
+    let (tasks, _) = index_scheduler
+        .get_task_ids_from_authorized_indexes(&query, &AuthFilter::default())
+        .unwrap();
+    // same query but with a valid uid
+    snapshot!(snapshot_bitmap(&tasks), @"[2,]");
+}
+
+#[test]
+fn query_tasks_special_rules() {
+    let (index_scheduler, mut handle) =
+        IndexScheduler::test(true, vec![(3, FailureLocation::InsideProcessBatch)]);
+
+    let kind = index_creation_task("catto", "mouse");
+    let _task = index_scheduler.register(kind, None, false).unwrap();
+    let kind = index_creation_task("doggo", "sheep");
+    let _task = index_scheduler.register(kind, None, false).unwrap();
+    let kind = KindWithContent::IndexSwap {
+        swaps: vec![IndexSwap { indexes: ("catto".to_owned(), "doggo".to_owned()) }],
+    };
+    let _task = index_scheduler.register(kind, None, false).unwrap();
+    let kind = KindWithContent::IndexSwap {
+        swaps: vec![IndexSwap { indexes: ("catto".to_owned(), "whalo".to_owned()) }],
+    };
+    let _task = index_scheduler.register(kind, None, false).unwrap();
+
+    snapshot!(snapshot_index_scheduler(&index_scheduler), name: "start");
+
+    handle.advance_till([Start, BatchCreated]);
+
+    let rtxn = index_scheduler.env.read_txn().unwrap();
+    let proc = index_scheduler.processing_tasks.read().unwrap();
+
+    let query = Query { index_uids: Some(vec!["catto".to_owned()]), ..Default::default() };
+    let (tasks, _) = index_scheduler
+        .queue
+        .get_task_ids_from_authorized_indexes(&rtxn, &query, &AuthFilter::default(), &proc)
+        .unwrap();
+    // only the first task associated with catto is returned, the indexSwap tasks are excluded!
+    snapshot!(snapshot_bitmap(&tasks), @"[0,]");
+
+    let query = Query { index_uids: Some(vec!["catto".to_owned()]), ..Default::default() };
+    let (tasks, _) = index_scheduler
+        .queue
+        .get_task_ids_from_authorized_indexes(
+            &rtxn,
+            &query,
+            &AuthFilter::with_allowed_indexes(
+                vec![IndexUidPattern::new_unchecked("doggo")].into_iter().collect(),
+            ),
+            &proc,
+        )
+        .unwrap();
+    // we have asked for only the tasks associated with catto, but are only authorized to retrieve the tasks
+    // associated with doggo -> empty result
+    snapshot!(snapshot_bitmap(&tasks), @"[]");
+
+    let query = Query::default();
+    let (tasks, _) = index_scheduler
+        .queue
+        .get_task_ids_from_authorized_indexes(
+            &rtxn,
+            &query,
+            &AuthFilter::with_allowed_indexes(
+                vec![IndexUidPattern::new_unchecked("doggo")].into_iter().collect(),
+            ),
+            &proc,
+        )
+        .unwrap();
+    // we asked for all the tasks, but we are only authorized to retrieve the doggo tasks
+    // -> only the index creation of doggo should be returned
+    snapshot!(snapshot_bitmap(&tasks), @"[1,]");
+
+    let query = Query::default();
+    let (tasks, _) = index_scheduler
+        .queue
+        .get_task_ids_from_authorized_indexes(
+            &rtxn,
+            &query,
+            &AuthFilter::with_allowed_indexes(
+                vec![
+                    IndexUidPattern::new_unchecked("catto"),
+                    IndexUidPattern::new_unchecked("doggo"),
+                ]
+                .into_iter()
+                .collect(),
+            ),
+            &proc,
+        )
+        .unwrap();
+    // we asked for all the tasks, but we are only authorized to retrieve the doggo and catto tasks
+    // -> all tasks except the swap of catto with whalo are returned
+    snapshot!(snapshot_bitmap(&tasks), @"[0,1,]");
+
+    let query = Query::default();
+    let (tasks, _) = index_scheduler
+        .queue
+        .get_task_ids_from_authorized_indexes(&rtxn, &query, &AuthFilter::default(), &proc)
+        .unwrap();
+    // we asked for all the tasks with all index authorized -> all tasks returned
+    snapshot!(snapshot_bitmap(&tasks), @"[0,1,2,3,]");
+}
+
+#[test]
+fn query_tasks_canceled_by() {
+    let (index_scheduler, mut handle) =
+        IndexScheduler::test(true, vec![(3, FailureLocation::InsideProcessBatch)]);
+
+    let kind = index_creation_task("catto", "mouse");
+    let _ = index_scheduler.register(kind, None, false).unwrap();
+    let kind = index_creation_task("doggo", "sheep");
+    let _ = index_scheduler.register(kind, None, false).unwrap();
+    let kind = KindWithContent::IndexSwap {
+        swaps: vec![IndexSwap { indexes: ("catto".to_owned(), "doggo".to_owned()) }],
+    };
+    let _task = index_scheduler.register(kind, None, false).unwrap();
+
+    handle.advance_n_successful_batches(1);
+    let kind = KindWithContent::TaskCancelation {
+        query: "test_query".to_string(),
+        tasks: [0, 1, 2, 3].into_iter().collect(),
+    };
+    let task_cancelation = index_scheduler.register(kind, None, false).unwrap();
+    handle.advance_n_successful_batches(1);
+
+    snapshot!(snapshot_index_scheduler(&index_scheduler), name: "start");
+
+    let rtxn = index_scheduler.read_txn().unwrap();
+    let proc = index_scheduler.processing_tasks.read().unwrap();
+    let query = Query { canceled_by: Some(vec![task_cancelation.uid]), ..Query::default() };
+    let (tasks, _) = index_scheduler
+        .queue
+        .get_task_ids_from_authorized_indexes(&rtxn, &query, &AuthFilter::default(), &proc)
+        .unwrap();
+    // 0 is not returned because it was not canceled, 3 is not returned because it is the uid of the
+    // taskCancelation itself
+    snapshot!(snapshot_bitmap(&tasks), @"[1,2,]");
+
+    let query = Query { canceled_by: Some(vec![task_cancelation.uid]), ..Query::default() };
+    let (tasks, _) = index_scheduler
+        .queue
+        .get_task_ids_from_authorized_indexes(
+            &rtxn,
+            &query,
+            &AuthFilter::with_allowed_indexes(
+                vec![IndexUidPattern::new_unchecked("doggo")].into_iter().collect(),
+            ),
+            &proc,
+        )
+        .unwrap();
+    // Return only 1 because the user is not authorized to see task 2
+    snapshot!(snapshot_bitmap(&tasks), @"[1,]");
+}

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

@@ -0,0 +1,398 @@
+use big_s::S;
+use meili_snap::{json_string, snapshot};
+use meilisearch_types::error::ErrorCode;
+use meilisearch_types::tasks::{KindWithContent, Status};
+use roaring::RoaringBitmap;
+
+use crate::insta_snapshot::snapshot_index_scheduler;
+use crate::test_utils::Breakpoint::*;
+use crate::test_utils::{index_creation_task, replace_document_import_task};
+use crate::{IndexScheduler, Query};
+
+#[test]
+fn register() {
+    // In this test, the handle doesn't make any progress, we only check that the tasks are registered
+    let (index_scheduler, mut _handle) = IndexScheduler::test(true, vec![]);
+
+    let kinds = [
+        index_creation_task("catto", "mouse"),
+        replace_document_import_task("catto", None, 0, 12),
+        replace_document_import_task("catto", None, 1, 50),
+        replace_document_import_task("doggo", Some("bone"), 2, 5000),
+    ];
+    let (_, file) = index_scheduler.queue.create_update_file_with_uuid(0).unwrap();
+    file.persist().unwrap();
+    let (_, file) = index_scheduler.queue.create_update_file_with_uuid(1).unwrap();
+    file.persist().unwrap();
+    let (_, file) = index_scheduler.queue.create_update_file_with_uuid(2).unwrap();
+    file.persist().unwrap();
+
+    for (idx, kind) in kinds.into_iter().enumerate() {
+        let k = kind.as_kind();
+        let task = index_scheduler.register(kind, None, false).unwrap();
+        index_scheduler.assert_internally_consistent();
+
+        assert_eq!(task.uid, idx as u32);
+        assert_eq!(task.status, Status::Enqueued);
+        assert_eq!(task.kind.as_kind(), k);
+    }
+
+    snapshot!(snapshot_index_scheduler(&index_scheduler), name: "everything_is_successfully_registered");
+}
+
+#[test]
+fn dry_run() {
+    let (index_scheduler, _handle) = IndexScheduler::test(true, vec![]);
+
+    let kind = KindWithContent::IndexCreation { index_uid: S("doggo"), primary_key: None };
+    let task = index_scheduler.register(kind, None, true).unwrap();
+    snapshot!(task.uid, @"0");
+    snapshot!(snapshot_index_scheduler(&index_scheduler), @r"
+        ### Autobatching Enabled = true
+        ### Processing batch None:
+        []
+        ----------------------------------------------------------------------
+        ### All Tasks:
+        ----------------------------------------------------------------------
+        ### Status:
+        ----------------------------------------------------------------------
+        ### Kind:
+        ----------------------------------------------------------------------
+        ### Index Tasks:
+        ----------------------------------------------------------------------
+        ### Index Mapper:
+
+        ----------------------------------------------------------------------
+        ### Canceled By:
+
+        ----------------------------------------------------------------------
+        ### Enqueued At:
+        ----------------------------------------------------------------------
+        ### Started At:
+        ----------------------------------------------------------------------
+        ### Finished At:
+        ----------------------------------------------------------------------
+        ### All Batches:
+        ----------------------------------------------------------------------
+        ### Batch to tasks mapping:
+        ----------------------------------------------------------------------
+        ### Batches Status:
+        ----------------------------------------------------------------------
+        ### Batches Kind:
+        ----------------------------------------------------------------------
+        ### Batches Index Tasks:
+        ----------------------------------------------------------------------
+        ### Batches Enqueued At:
+        ----------------------------------------------------------------------
+        ### Batches Started At:
+        ----------------------------------------------------------------------
+        ### Batches Finished At:
+        ----------------------------------------------------------------------
+        ### File Store:
+
+        ----------------------------------------------------------------------
+        ");
+
+    let kind = KindWithContent::IndexCreation { index_uid: S("doggo"), primary_key: None };
+    let task = index_scheduler.register(kind, Some(12), true).unwrap();
+    snapshot!(task.uid, @"12");
+    snapshot!(snapshot_index_scheduler(&index_scheduler), @r"
+        ### Autobatching Enabled = true
+        ### Processing batch None:
+        []
+        ----------------------------------------------------------------------
+        ### All Tasks:
+        ----------------------------------------------------------------------
+        ### Status:
+        ----------------------------------------------------------------------
+        ### Kind:
+        ----------------------------------------------------------------------
+        ### Index Tasks:
+        ----------------------------------------------------------------------
+        ### Index Mapper:
+
+        ----------------------------------------------------------------------
+        ### Canceled By:
+
+        ----------------------------------------------------------------------
+        ### Enqueued At:
+        ----------------------------------------------------------------------
+        ### Started At:
+        ----------------------------------------------------------------------
+        ### Finished At:
+        ----------------------------------------------------------------------
+        ### All Batches:
+        ----------------------------------------------------------------------
+        ### Batch to tasks mapping:
+        ----------------------------------------------------------------------
+        ### Batches Status:
+        ----------------------------------------------------------------------
+        ### Batches Kind:
+        ----------------------------------------------------------------------
+        ### Batches Index Tasks:
+        ----------------------------------------------------------------------
+        ### Batches Enqueued At:
+        ----------------------------------------------------------------------
+        ### Batches Started At:
+        ----------------------------------------------------------------------
+        ### Batches Finished At:
+        ----------------------------------------------------------------------
+        ### File Store:
+
+        ----------------------------------------------------------------------
+        ");
+}
+
+#[test]
+fn basic_set_taskid() {
+    let (index_scheduler, _handle) = IndexScheduler::test(true, vec![]);
+
+    let kind = KindWithContent::IndexCreation { index_uid: S("doggo"), primary_key: None };
+    let task = index_scheduler.register(kind, None, false).unwrap();
+    snapshot!(task.uid, @"0");
+
+    let kind = KindWithContent::IndexCreation { index_uid: S("doggo"), primary_key: None };
+    let task = index_scheduler.register(kind, Some(12), false).unwrap();
+    snapshot!(task.uid, @"12");
+
+    let kind = KindWithContent::IndexCreation { index_uid: S("doggo"), primary_key: None };
+    let error = index_scheduler.register(kind, Some(5), false).unwrap_err();
+    snapshot!(error, @"Received bad task id: 5 should be >= to 13.");
+}
+
+#[test]
+fn test_disable_auto_deletion_of_tasks() {
+    let (index_scheduler, mut handle) = IndexScheduler::test_with_custom_config(vec![], |config| {
+        config.cleanup_enabled = false;
+        config.max_number_of_tasks = 2;
+        None
+    });
+
+    index_scheduler
+        .register(
+            KindWithContent::IndexCreation { index_uid: S("doggo"), primary_key: None },
+            None,
+            false,
+        )
+        .unwrap();
+    handle.advance_one_successful_batch();
+
+    index_scheduler
+        .register(
+            KindWithContent::IndexCreation { index_uid: S("doggo"), primary_key: None },
+            None,
+            false,
+        )
+        .unwrap();
+    handle.advance_one_failed_batch();
+
+    // at this point the max number of tasks is reached
+    // we can still enqueue multiple tasks
+    index_scheduler
+        .register(
+            KindWithContent::IndexCreation { index_uid: S("doggo"), primary_key: None },
+            None,
+            false,
+        )
+        .unwrap();
+    index_scheduler
+        .register(
+            KindWithContent::IndexCreation { index_uid: S("doggo"), primary_key: None },
+            None,
+            false,
+        )
+        .unwrap();
+
+    let rtxn = index_scheduler.env.read_txn().unwrap();
+    let proc = index_scheduler.processing_tasks.read().unwrap();
+    let tasks =
+        index_scheduler.queue.get_task_ids(&rtxn, &Query { ..Default::default() }, &proc).unwrap();
+    let tasks = index_scheduler.queue.tasks.get_existing_tasks(&rtxn, tasks).unwrap();
+    snapshot!(json_string!(tasks, { "[].enqueuedAt" => "[date]", "[].startedAt" => "[date]", "[].finishedAt" => "[date]" }), name: "task_queue_is_full");
+    drop(rtxn);
+    drop(proc);
+
+    // now we're above the max number of tasks
+    // and if we try to advance in the tick function no new task deletion should be enqueued
+    handle.advance_till([Start, BatchCreated]);
+    let rtxn = index_scheduler.env.read_txn().unwrap();
+    let proc = index_scheduler.processing_tasks.read().unwrap();
+    let tasks =
+        index_scheduler.queue.get_task_ids(&rtxn, &Query { ..Default::default() }, &proc).unwrap();
+    let tasks = index_scheduler.queue.tasks.get_existing_tasks(&rtxn, tasks).unwrap();
+    snapshot!(json_string!(tasks, { "[].enqueuedAt" => "[date]", "[].startedAt" => "[date]", "[].finishedAt" => "[date]", ".**.original_filter" => "[filter]", ".**.query" => "[query]" }), name: "task_deletion_have_not_been_enqueued");
+    drop(rtxn);
+    drop(proc);
+}
+
+#[test]
+fn test_auto_deletion_of_tasks() {
+    let (index_scheduler, mut handle) = IndexScheduler::test_with_custom_config(vec![], |config| {
+        config.max_number_of_tasks = 2;
+        None
+    });
+
+    index_scheduler
+        .register(
+            KindWithContent::IndexCreation { index_uid: S("doggo"), primary_key: None },
+            None,
+            false,
+        )
+        .unwrap();
+    handle.advance_one_successful_batch();
+
+    index_scheduler
+        .register(
+            KindWithContent::IndexCreation { index_uid: S("doggo"), primary_key: None },
+            None,
+            false,
+        )
+        .unwrap();
+    handle.advance_one_failed_batch();
+
+    // at this point the max number of tasks is reached
+    // we can still enqueue multiple tasks
+    index_scheduler
+        .register(
+            KindWithContent::IndexCreation { index_uid: S("doggo"), primary_key: None },
+            None,
+            false,
+        )
+        .unwrap();
+    index_scheduler
+        .register(
+            KindWithContent::IndexCreation { index_uid: S("doggo"), primary_key: None },
+            None,
+            false,
+        )
+        .unwrap();
+
+    let rtxn = index_scheduler.env.read_txn().unwrap();
+    let proc = index_scheduler.processing_tasks.read().unwrap();
+    let tasks =
+        index_scheduler.queue.get_task_ids(&rtxn, &Query { ..Default::default() }, &proc).unwrap();
+    let tasks = index_scheduler.queue.tasks.get_existing_tasks(&rtxn, tasks).unwrap();
+    snapshot!(json_string!(tasks, { "[].enqueuedAt" => "[date]", "[].startedAt" => "[date]", "[].finishedAt" => "[date]" }), name: "task_queue_is_full");
+    drop(rtxn);
+    drop(proc);
+
+    // now we're above the max number of tasks
+    // and if we try to advance in the tick function a new task deletion should be enqueued
+    handle.advance_till([Start, BatchCreated]);
+    let rtxn = index_scheduler.env.read_txn().unwrap();
+    let proc = index_scheduler.processing_tasks.read().unwrap();
+    let tasks =
+        index_scheduler.queue.get_task_ids(&rtxn, &Query { ..Default::default() }, &proc).unwrap();
+    let tasks = index_scheduler.queue.tasks.get_existing_tasks(&rtxn, tasks).unwrap();
+    snapshot!(json_string!(tasks, { "[].enqueuedAt" => "[date]", "[].startedAt" => "[date]", "[].finishedAt" => "[date]", ".**.original_filter" => "[filter]", ".**.query" => "[query]" }), name: "task_deletion_have_been_enqueued");
+    drop(rtxn);
+    drop(proc);
+
+    handle.advance_till([InsideProcessBatch, ProcessBatchSucceeded, AfterProcessing]);
+    let rtxn = index_scheduler.env.read_txn().unwrap();
+    let proc = index_scheduler.processing_tasks.read().unwrap();
+    let tasks =
+        index_scheduler.queue.get_task_ids(&rtxn, &Query { ..Default::default() }, &proc).unwrap();
+    let tasks = index_scheduler.queue.tasks.get_existing_tasks(&rtxn, tasks).unwrap();
+    snapshot!(json_string!(tasks, { "[].enqueuedAt" => "[date]", "[].startedAt" => "[date]", "[].finishedAt" => "[date]", ".**.original_filter" => "[filter]", ".**.query" => "[query]" }), name: "task_deletion_have_been_processed");
+    drop(rtxn);
+    drop(proc);
+
+    handle.advance_one_failed_batch();
+    // a new task deletion has been enqueued
+    handle.advance_one_successful_batch();
+    let rtxn = index_scheduler.env.read_txn().unwrap();
+    let proc = index_scheduler.processing_tasks.read().unwrap();
+    let tasks =
+        index_scheduler.queue.get_task_ids(&rtxn, &Query { ..Default::default() }, &proc).unwrap();
+    let tasks = index_scheduler.queue.tasks.get_existing_tasks(&rtxn, tasks).unwrap();
+    snapshot!(json_string!(tasks, { "[].enqueuedAt" => "[date]", "[].startedAt" => "[date]", "[].finishedAt" => "[date]", ".**.original_filter" => "[filter]", ".**.query" => "[query]" }), name: "after_the_second_task_deletion");
+    drop(rtxn);
+    drop(proc);
+
+    handle.advance_one_failed_batch();
+    handle.advance_one_successful_batch();
+    let rtxn = index_scheduler.env.read_txn().unwrap();
+    let proc = index_scheduler.processing_tasks.read().unwrap();
+    let tasks =
+        index_scheduler.queue.get_task_ids(&rtxn, &Query { ..Default::default() }, &proc).unwrap();
+    let tasks = index_scheduler.queue.tasks.get_existing_tasks(&rtxn, tasks).unwrap();
+    snapshot!(json_string!(tasks, { "[].enqueuedAt" => "[date]", "[].startedAt" => "[date]", "[].finishedAt" => "[date]", ".**.original_filter" => "[filter]", ".**.query" => "[query]" }), name: "everything_has_been_processed");
+    drop(rtxn);
+    drop(proc);
+}
+
+#[test]
+fn test_task_queue_is_full() {
+    let (index_scheduler, mut handle) = IndexScheduler::test_with_custom_config(vec![], |config| {
+        // that's the minimum map size possible
+        config.task_db_size = 1048576 * 3;
+        None
+    });
+
+    index_scheduler
+        .register(
+            KindWithContent::IndexCreation { index_uid: S("doggo"), primary_key: None },
+            None,
+            false,
+        )
+        .unwrap();
+    handle.advance_one_successful_batch();
+    // on average this task takes ~600 bytes
+    loop {
+        let result = index_scheduler.register(
+            KindWithContent::IndexCreation { index_uid: S("doggo"), primary_key: None },
+            None,
+            false,
+        );
+        if result.is_err() {
+            break;
+        }
+        handle.advance_one_failed_batch();
+    }
+    index_scheduler.assert_internally_consistent();
+
+    // at this point the task DB shoud have reached its limit and we should not be able to register new tasks
+    let result = index_scheduler
+        .register(
+            KindWithContent::IndexCreation { index_uid: S("doggo"), primary_key: None },
+            None,
+            false,
+        )
+        .unwrap_err();
+    snapshot!(result, @"Meilisearch cannot receive write operations because the limit of the task database has been reached. Please delete tasks to continue performing write operations.");
+    // we won't be able to test this error in an integration test thus as a best effort test I still ensure the error return the expected error code
+    snapshot!(format!("{:?}", result.error_code()), @"NoSpaceLeftOnDevice");
+
+    // Even the task deletion that doesn't delete anything shouldn't be accepted
+    let result = index_scheduler
+        .register(
+            KindWithContent::TaskDeletion { query: S("test"), tasks: RoaringBitmap::new() },
+            None,
+            false,
+        )
+        .unwrap_err();
+    snapshot!(result, @"Meilisearch cannot receive write operations because the limit of the task database has been reached. Please delete tasks to continue performing write operations.");
+    // we won't be able to test this error in an integration test thus as a best effort test I still ensure the error return the expected error code
+    snapshot!(format!("{:?}", result.error_code()), @"NoSpaceLeftOnDevice");
+
+    // But a task deletion that delete something should works
+    index_scheduler
+        .register(
+            KindWithContent::TaskDeletion { query: S("test"), tasks: (0..100).collect() },
+            None,
+            false,
+        )
+        .unwrap();
+    handle.advance_one_successful_batch();
+
+    // Now we should be able to enqueue a few tasks again
+    index_scheduler
+        .register(
+            KindWithContent::IndexCreation { index_uid: S("doggo"), primary_key: None },
+            None,
+            false,
+        )
+        .unwrap();
+    handle.advance_one_failed_batch();
+}

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

@@ -0,0 +1,474 @@
+/*!
+The autobatcher is responsible for combining the next enqueued
+tasks affecting a single index into a [batch](crate::batch::Batch).
+
+The main function of the autobatcher is [`next_autobatch`].
+*/
+
+use meilisearch_types::tasks::TaskId;
+use std::ops::ControlFlow::{self, Break, Continue};
+
+use crate::KindWithContent;
+
+/// Succinctly describes a task's [`Kind`](meilisearch_types::tasks::Kind)
+/// for the purpose of simplifying the implementation of the autobatcher.
+///
+/// Only the non-prioritised tasks that can be grouped in a batch have a corresponding [`AutobatchKind`]
+enum AutobatchKind {
+    DocumentImport { allow_index_creation: bool, primary_key: Option<String> },
+    DocumentEdition,
+    DocumentDeletion { by_filter: bool },
+    DocumentClear,
+    Settings { allow_index_creation: bool },
+    IndexCreation,
+    IndexDeletion,
+    IndexUpdate,
+    IndexSwap,
+}
+
+impl AutobatchKind {
+    #[rustfmt::skip]
+    fn allow_index_creation(&self) -> Option<bool> {
+        match self {
+            AutobatchKind::DocumentImport { allow_index_creation, .. }
+            | AutobatchKind::Settings { allow_index_creation, .. } => Some(*allow_index_creation),
+            _ => None,
+        }
+    }
+
+    fn primary_key(&self) -> Option<Option<&str>> {
+        match self {
+            AutobatchKind::DocumentImport { primary_key, .. } => Some(primary_key.as_deref()),
+            _ => None,
+        }
+    }
+}
+
+impl From<KindWithContent> for AutobatchKind {
+    fn from(kind: KindWithContent) -> Self {
+        match kind {
+            KindWithContent::DocumentAdditionOrUpdate {
+                allow_index_creation, primary_key, ..
+            } => AutobatchKind::DocumentImport { allow_index_creation, primary_key },
+            KindWithContent::DocumentEdition { .. } => AutobatchKind::DocumentEdition,
+            KindWithContent::DocumentDeletion { .. } => {
+                AutobatchKind::DocumentDeletion { by_filter: false }
+            }
+            KindWithContent::DocumentClear { .. } => AutobatchKind::DocumentClear,
+            KindWithContent::DocumentDeletionByFilter { .. } => {
+                AutobatchKind::DocumentDeletion { by_filter: true }
+            }
+            KindWithContent::SettingsUpdate { allow_index_creation, is_deletion, .. } => {
+                AutobatchKind::Settings {
+                    allow_index_creation: allow_index_creation && !is_deletion,
+                }
+            }
+            KindWithContent::IndexDeletion { .. } => AutobatchKind::IndexDeletion,
+            KindWithContent::IndexCreation { .. } => AutobatchKind::IndexCreation,
+            KindWithContent::IndexUpdate { .. } => AutobatchKind::IndexUpdate,
+            KindWithContent::IndexSwap { .. } => AutobatchKind::IndexSwap,
+            KindWithContent::TaskCancelation { .. }
+            | KindWithContent::TaskDeletion { .. }
+            | KindWithContent::DumpCreation { .. }
+            | KindWithContent::UpgradeDatabase { .. }
+            | KindWithContent::SnapshotCreation => {
+                panic!("The autobatcher should never be called with tasks that don't apply to an index.")
+            }
+        }
+    }
+}
+
+#[derive(Debug)]
+pub enum BatchKind {
+    DocumentClear {
+        ids: Vec<TaskId>,
+    },
+    DocumentOperation {
+        allow_index_creation: bool,
+        primary_key: Option<String>,
+        operation_ids: Vec<TaskId>,
+    },
+    DocumentEdition {
+        id: TaskId,
+    },
+    DocumentDeletion {
+        deletion_ids: Vec<TaskId>,
+        includes_by_filter: bool,
+    },
+    ClearAndSettings {
+        other: Vec<TaskId>,
+        allow_index_creation: bool,
+        settings_ids: Vec<TaskId>,
+    },
+    Settings {
+        allow_index_creation: bool,
+        settings_ids: Vec<TaskId>,
+    },
+    IndexDeletion {
+        ids: Vec<TaskId>,
+    },
+    IndexCreation {
+        id: TaskId,
+    },
+    IndexUpdate {
+        id: TaskId,
+    },
+    IndexSwap {
+        id: TaskId,
+    },
+}
+
+impl BatchKind {
+    #[rustfmt::skip]
+    fn allow_index_creation(&self) -> Option<bool> {
+        match self {
+            BatchKind::DocumentOperation { allow_index_creation, .. }
+            | BatchKind::ClearAndSettings { allow_index_creation, .. }
+            | BatchKind::Settings { allow_index_creation, .. } => Some(*allow_index_creation),
+            _ => None,
+        }
+    }
+
+    fn primary_key(&self) -> Option<Option<&str>> {
+        match self {
+            BatchKind::DocumentOperation { primary_key, .. } => Some(primary_key.as_deref()),
+            _ => None,
+        }
+    }
+}
+
+impl BatchKind {
+    /// Returns a `ControlFlow::Break` if you must stop right now.
+    /// The boolean tell you if an index has been created by the batched task.
+    /// To ease the writing of the code. `true` can be returned when you don't need to create an index
+    /// but false can't be returned if you needs to create an index.
+    // TODO use an AutoBatchKind as input
+    pub fn new(
+        task_id: TaskId,
+        kind: KindWithContent,
+        primary_key: Option<&str>,
+    ) -> (ControlFlow<BatchKind, BatchKind>, bool) {
+        use AutobatchKind as K;
+
+        match AutobatchKind::from(kind) {
+            K::IndexCreation => (Break(BatchKind::IndexCreation { id: task_id }), true),
+            K::IndexDeletion => (Break(BatchKind::IndexDeletion { ids: vec![task_id] }), false),
+            K::IndexUpdate => (Break(BatchKind::IndexUpdate { id: task_id }), false),
+            K::IndexSwap => (Break(BatchKind::IndexSwap { id: task_id }), false),
+            K::DocumentClear => (Continue(BatchKind::DocumentClear { ids: vec![task_id] }), false),
+            K::DocumentImport { allow_index_creation, primary_key: pk }
+                if primary_key.is_none() || pk.is_none() || primary_key == pk.as_deref() =>
+            {
+                (
+                    Continue(BatchKind::DocumentOperation {
+                        allow_index_creation,
+                        primary_key: pk,
+                        operation_ids: vec![task_id],
+                    }),
+                    allow_index_creation,
+                )
+            }
+            // if the primary key set in the task was different than ours we should stop and make this batch fail asap.
+            K::DocumentImport { allow_index_creation, primary_key } => (
+                Break(BatchKind::DocumentOperation {
+                    allow_index_creation,
+                    primary_key,
+                    operation_ids: vec![task_id],
+                }),
+                allow_index_creation,
+            ),
+            K::DocumentEdition => (Break(BatchKind::DocumentEdition { id: task_id }), false),
+            K::DocumentDeletion { by_filter: includes_by_filter } => (
+                Continue(BatchKind::DocumentDeletion {
+                    deletion_ids: vec![task_id],
+                    includes_by_filter,
+                }),
+                false,
+            ),
+            K::Settings { allow_index_creation } => (
+                Continue(BatchKind::Settings { allow_index_creation, settings_ids: vec![task_id] }),
+                allow_index_creation,
+            ),
+        }
+    }
+
+    /// Returns a `ControlFlow::Break` if you must stop right now.
+    /// The boolean tell you if an index has been created by the batched task.
+    /// To ease the writing of the code. `true` can be returned when you don't need to create an index
+    /// but false can't be returned if you needs to create an index.
+    #[rustfmt::skip]
+    fn accumulate(self, id: TaskId, kind: AutobatchKind, index_already_exists: bool, primary_key: Option<&str>) -> ControlFlow<BatchKind, BatchKind> {
+        use AutobatchKind as K;
+
+        match (self, kind) {
+            // We don't batch any of these operations
+            (this, K::IndexCreation | K::IndexUpdate | K::IndexSwap | K::DocumentEdition) => Break(this),
+            // We must not batch tasks that don't have the same index creation rights if the index doesn't already exists.
+            (this, kind) if !index_already_exists && this.allow_index_creation() == Some(false) && kind.allow_index_creation() == Some(true) => {
+                Break(this)
+            },
+            // NOTE: We need to negate the whole condition since we're checking if we need to break instead of continue.
+            //       I wrote it this way because it's easier to understand than the other way around.
+            (this, kind) if !(
+                // 1. If both task don't interact with primary key -> we can continue
+                (this.primary_key().is_none() && kind.primary_key().is_none()) ||
+                // 2. Else ->
+                (
+                    // 2.1 If we already have a primary-key ->
+                    (
+                        primary_key.is_some() &&
+                        // 2.1.1 If the task we're trying to accumulate have a pk it must be equal to our primary key
+                        // 2.1.2 If the task don't have a primary-key -> we can continue
+                        kind.primary_key().map_or(true, |pk| pk == primary_key)
+                    ) ||
+                    // 2.2 If we don't have a primary-key ->
+                    (
+                        // 2.2.1 If both the batch and the task have a primary key they should be equal
+                        // 2.2.2 If the batch is set to Some(None), the task should be too
+                        // 2.2.3 If the batch is set to None -> we can continue
+                        this.primary_key().zip(kind.primary_key()).map_or(true, |(this, kind)| this == kind)
+                    )
+                )
+
+                ) // closing the negation
+
+            => {
+                Break(this)
+            },
+            // The index deletion can batch with everything but must stop after
+            (
+                BatchKind::DocumentClear { mut ids }
+                | BatchKind::DocumentDeletion { deletion_ids: mut ids, includes_by_filter: _ }
+                | BatchKind::DocumentOperation { allow_index_creation: _, primary_key: _, operation_ids: mut ids }
+                | BatchKind::Settings { allow_index_creation: _, settings_ids: mut ids },
+                K::IndexDeletion,
+            ) => {
+                ids.push(id);
+                Break(BatchKind::IndexDeletion { ids })
+            }
+            (
+                BatchKind::ClearAndSettings { settings_ids: mut ids, allow_index_creation: _, mut other },
+                K::IndexDeletion,
+            ) => {
+                ids.push(id);
+                ids.append(&mut other);
+                Break(BatchKind::IndexDeletion { ids })
+            }
+
+            (
+                BatchKind::DocumentClear { mut ids },
+                K::DocumentClear | K::DocumentDeletion { by_filter: _ },
+            ) => {
+                ids.push(id);
+                Continue(BatchKind::DocumentClear { ids })
+            }
+            (
+                this @ BatchKind::DocumentClear { .. },
+                K::DocumentImport { .. } | K::Settings { .. },
+            ) => Break(this),
+            (
+                BatchKind::DocumentOperation { allow_index_creation: _, primary_key: _, mut operation_ids },
+                K::DocumentClear,
+            ) => {
+                operation_ids.push(id);
+                Continue(BatchKind::DocumentClear { ids: operation_ids })
+            }
+
+            // we can autobatch different kind of document operations and mix replacements with updates
+            (
+                BatchKind::DocumentOperation { allow_index_creation, primary_key: _, mut operation_ids },
+                K::DocumentImport { primary_key: pk, .. },
+            ) => {
+                operation_ids.push(id);
+                Continue(BatchKind::DocumentOperation {
+                    allow_index_creation,
+                    operation_ids,
+                    primary_key: pk,
+                })
+            }
+            (
+                BatchKind::DocumentOperation { allow_index_creation, primary_key, mut operation_ids },
+                K::DocumentDeletion { by_filter: false },
+            ) => {
+                operation_ids.push(id);
+
+                Continue(BatchKind::DocumentOperation {
+                    allow_index_creation,
+                    primary_key,
+                    operation_ids,
+                })
+            }
+            // We can't batch a document operation with a delete by filter
+            (
+                this @ BatchKind::DocumentOperation { .. },
+                K::DocumentDeletion { by_filter: true },
+            ) => {
+                Break(this)
+            }
+            (
+                this @ BatchKind::DocumentOperation { .. },
+                K::Settings { .. },
+            ) => Break(this),
+
+            (BatchKind::DocumentDeletion { mut deletion_ids, includes_by_filter: _ }, K::DocumentClear) => {
+                deletion_ids.push(id);
+                Continue(BatchKind::DocumentClear { ids: deletion_ids })
+            }
+            // we can't autobatch the deletion and import if the document deletion contained a filter
+            (
+                this @ BatchKind::DocumentDeletion { deletion_ids: _, includes_by_filter: true },
+                K::DocumentImport { .. }
+            ) => Break(this),
+            // we can autobatch the deletion and import if the index already exists
+            (
+                BatchKind::DocumentDeletion { mut deletion_ids, includes_by_filter: false },
+                K::DocumentImport { allow_index_creation, primary_key }
+            ) if index_already_exists => {
+                deletion_ids.push(id);
+
+                Continue(BatchKind::DocumentOperation {
+                    allow_index_creation,
+                    primary_key,
+                    operation_ids: deletion_ids,
+                })
+            }
+            // we can autobatch the deletion and import if both can't create an index
+            (
+                BatchKind::DocumentDeletion { mut deletion_ids, includes_by_filter: false },
+                K::DocumentImport { allow_index_creation, primary_key }
+            ) if !allow_index_creation => {
+                deletion_ids.push(id);
+
+                Continue(BatchKind::DocumentOperation {
+                    allow_index_creation,
+                    primary_key,
+                    operation_ids: deletion_ids,
+                })
+            }
+            // we can't autobatch a deletion and an import if the index does not exists but would be created by an addition
+            (
+                this @ BatchKind::DocumentDeletion { .. },
+                K::DocumentImport { .. }
+            ) => {
+                Break(this)
+            }
+            (BatchKind::DocumentDeletion { mut deletion_ids, includes_by_filter }, K::DocumentDeletion { by_filter }) => {
+                deletion_ids.push(id);
+                Continue(BatchKind::DocumentDeletion { deletion_ids, includes_by_filter: includes_by_filter | by_filter })
+            }
+            (this @ BatchKind::DocumentDeletion { .. }, K::Settings { .. }) => Break(this),
+
+            (
+                BatchKind::Settings { settings_ids, allow_index_creation },
+                K::DocumentClear,
+            ) => Continue(BatchKind::ClearAndSettings {
+                settings_ids,
+                allow_index_creation,
+                other: vec![id],
+            }),
+            (
+                this @ BatchKind::Settings { .. },
+                K::DocumentImport { .. } | K::DocumentDeletion { .. },
+            ) => Break(this),
+            (
+                BatchKind::Settings { mut settings_ids, allow_index_creation },
+                K::Settings { .. },
+            ) => {
+                settings_ids.push(id);
+                Continue(BatchKind::Settings {
+                    allow_index_creation,
+                    settings_ids,
+                })
+            }
+
+            (
+                BatchKind::ClearAndSettings { mut other, settings_ids, allow_index_creation },
+                K::DocumentClear,
+            ) => {
+                other.push(id);
+                Continue(BatchKind::ClearAndSettings {
+                    other,
+                    settings_ids,
+                    allow_index_creation,
+                })
+            }
+            (this @ BatchKind::ClearAndSettings { .. }, K::DocumentImport { .. }) => Break(this),
+            (
+                BatchKind::ClearAndSettings {
+                    mut other,
+                    settings_ids,
+                    allow_index_creation,
+                },
+                K::DocumentDeletion { .. },
+            ) => {
+                other.push(id);
+                Continue(BatchKind::ClearAndSettings {
+                    other,
+                    settings_ids,
+                    allow_index_creation,
+                })
+            }
+            (
+                BatchKind::ClearAndSettings { mut settings_ids, other, allow_index_creation },
+                K::Settings { .. },
+            ) => {
+                settings_ids.push(id);
+                Continue(BatchKind::ClearAndSettings {
+                    other,
+                    settings_ids,
+                    allow_index_creation,
+                })
+            }
+
+            (
+                BatchKind::IndexCreation { .. }
+                | BatchKind::IndexDeletion { .. }
+                | BatchKind::IndexUpdate { .. }
+                | BatchKind::IndexSwap { .. }
+                | BatchKind::DocumentEdition { .. },
+                _,
+            ) => {
+                unreachable!()
+            }
+        }
+    }
+}
+
+/// Create a batch from an ordered list of tasks.
+///
+/// ## Preconditions
+/// 1. The tasks must be enqueued and given in the order in which they were enqueued
+/// 2. The tasks must not be prioritised tasks (e.g. task cancellation, dump, snapshot, task deletion)
+/// 3. The tasks must all be related to the same index
+///
+/// ## Return
+/// `None` if the list of tasks is empty. Otherwise, an [`AutoBatch`] that represents
+/// a subset of the given tasks.
+pub fn autobatch(
+    enqueued: Vec<(TaskId, KindWithContent)>,
+    index_already_exists: bool,
+    primary_key: Option<&str>,
+) -> Option<(BatchKind, bool)> {
+    let mut enqueued = enqueued.into_iter();
+    let (id, kind) = enqueued.next()?;
+
+    // index_exist will keep track of if the index should exist at this point after the tasks we batched.
+    let mut index_exist = index_already_exists;
+
+    let (mut acc, must_create_index) = match BatchKind::new(id, kind, primary_key) {
+        (Continue(acc), create) => (acc, create),
+        (Break(acc), create) => return Some((acc, create)),
+    };
+
+    // if an index has been created in the previous step we can consider it as existing.
+    index_exist |= must_create_index;
+
+    for (id, kind) in enqueued {
+        acc = match acc.accumulate(id, kind.into(), index_exist, primary_key) {
+            Continue(acc) => acc,
+            Break(acc) => return Some((acc, must_create_index)),
+        };
+    }
+
+    Some((acc, must_create_index))
+}

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

@@ -0,0 +1,395 @@
+use meilisearch_types::milli::update::IndexDocumentsMethod::{
+    self, ReplaceDocuments, UpdateDocuments,
+};
+use meilisearch_types::tasks::{IndexSwap, KindWithContent};
+use uuid::Uuid;
+
+use self::autobatcher::{autobatch, BatchKind};
+use super::*;
+use crate::TaskId;
+
+#[macro_export]
+macro_rules! debug_snapshot {
+        ($value:expr, @$snapshot:literal) => {{
+            let value = format!("{:?}", $value);
+            meili_snap::snapshot!(value, @$snapshot);
+        }};
+    }
+
+fn autobatch_from(
+    index_already_exists: bool,
+    primary_key: Option<&str>,
+    input: impl IntoIterator<Item = KindWithContent>,
+) -> Option<(BatchKind, bool)> {
+    autobatch(
+        input.into_iter().enumerate().map(|(id, kind)| (id as TaskId, kind)).collect(),
+        index_already_exists,
+        primary_key,
+    )
+}
+
+fn doc_imp(
+    method: IndexDocumentsMethod,
+    allow_index_creation: bool,
+    primary_key: Option<&str>,
+) -> KindWithContent {
+    KindWithContent::DocumentAdditionOrUpdate {
+        index_uid: String::from("doggo"),
+        primary_key: primary_key.map(|pk| pk.to_string()),
+        method,
+        content_file: Uuid::new_v4(),
+        documents_count: 0,
+        allow_index_creation,
+    }
+}
+
+fn doc_del() -> KindWithContent {
+    KindWithContent::DocumentDeletion {
+        index_uid: String::from("doggo"),
+        documents_ids: Vec::new(),
+    }
+}
+
+fn doc_del_fil() -> KindWithContent {
+    KindWithContent::DocumentDeletionByFilter {
+        index_uid: String::from("doggo"),
+        filter_expr: serde_json::json!("cuteness > 100"),
+    }
+}
+
+fn doc_clr() -> KindWithContent {
+    KindWithContent::DocumentClear { index_uid: String::from("doggo") }
+}
+
+fn settings(allow_index_creation: bool) -> KindWithContent {
+    KindWithContent::SettingsUpdate {
+        index_uid: String::from("doggo"),
+        new_settings: Default::default(),
+        is_deletion: false,
+        allow_index_creation,
+    }
+}
+
+fn idx_create() -> KindWithContent {
+    KindWithContent::IndexCreation { index_uid: String::from("doggo"), primary_key: None }
+}
+
+fn idx_update() -> KindWithContent {
+    KindWithContent::IndexUpdate { index_uid: String::from("doggo"), primary_key: None }
+}
+
+fn idx_del() -> KindWithContent {
+    KindWithContent::IndexDeletion { index_uid: String::from("doggo") }
+}
+
+fn idx_swap() -> KindWithContent {
+    KindWithContent::IndexSwap {
+        swaps: vec![IndexSwap { indexes: (String::from("doggo"), String::from("catto")) }],
+    }
+}
+
+#[test]
+fn autobatch_simple_operation_together() {
+    // we can autobatch one or multiple `ReplaceDocuments` together.
+    // if the index exists.
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, false, None)]), @"Some((DocumentOperation { allow_index_creation: false, primary_key: None, operation_ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None), doc_imp( ReplaceDocuments, true , None), doc_imp(ReplaceDocuments, true , None)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0, 1, 2] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, false, None), doc_imp( ReplaceDocuments, false , None), doc_imp(ReplaceDocuments, false , None)]), @"Some((DocumentOperation { allow_index_creation: false, primary_key: None, operation_ids: [0, 1, 2] }, false))");
+
+    // if it doesn't exists.
+    debug_snapshot!(autobatch_from(false,None,  [doc_imp(ReplaceDocuments, true, None)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(false,None,  [doc_imp(ReplaceDocuments, false, None)]), @"Some((DocumentOperation { allow_index_creation: false, primary_key: None, operation_ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(false,None,  [doc_imp(ReplaceDocuments, true, None), doc_imp( ReplaceDocuments, true , None), doc_imp(ReplaceDocuments, true , None)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0, 1, 2] }, true))");
+    debug_snapshot!(autobatch_from(false,None,  [doc_imp(ReplaceDocuments, false, None), doc_imp( ReplaceDocuments, true , None), doc_imp(ReplaceDocuments, true , None)]), @"Some((DocumentOperation { allow_index_creation: false, primary_key: None, operation_ids: [0] }, false))");
+
+    // we can autobatch one or multiple `UpdateDocuments` together.
+    // if the index exists.
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, true, None)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, true, None), doc_imp(UpdateDocuments, true, None), doc_imp(UpdateDocuments, true, None)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0, 1, 2] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, false, None)]), @"Some((DocumentOperation { allow_index_creation: false, primary_key: None, operation_ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, false, None), doc_imp(UpdateDocuments, false, None), doc_imp(UpdateDocuments, false, None)]), @"Some((DocumentOperation { allow_index_creation: false, primary_key: None, operation_ids: [0, 1, 2] }, false))");
+
+    // if it doesn't exists.
+    debug_snapshot!(autobatch_from(false,None,  [doc_imp(UpdateDocuments, true, None)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(false,None,  [doc_imp(UpdateDocuments, true, None), doc_imp(UpdateDocuments, true, None), doc_imp(UpdateDocuments, true, None)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0, 1, 2] }, true))");
+    debug_snapshot!(autobatch_from(false,None,  [doc_imp(UpdateDocuments, false, None)]), @"Some((DocumentOperation { allow_index_creation: false, primary_key: None, operation_ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(false,None,  [doc_imp(UpdateDocuments, false, None), doc_imp(UpdateDocuments, false, None), doc_imp(UpdateDocuments, false, None)]), @"Some((DocumentOperation { allow_index_creation: false, primary_key: None, operation_ids: [0, 1, 2] }, false))");
+
+    // we can autobatch one or multiple DocumentDeletion together
+    debug_snapshot!(autobatch_from(true, None, [doc_del()]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: false }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_del(), doc_del(), doc_del()]), @"Some((DocumentDeletion { deletion_ids: [0, 1, 2], includes_by_filter: false }, false))");
+    debug_snapshot!(autobatch_from(false,None,  [doc_del()]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: false }, false))");
+    debug_snapshot!(autobatch_from(false,None,  [doc_del(), doc_del(), doc_del()]), @"Some((DocumentDeletion { deletion_ids: [0, 1, 2], includes_by_filter: false }, false))");
+
+    // we can autobatch one or multiple DocumentDeletionByFilter together
+    debug_snapshot!(autobatch_from(true, None, [doc_del_fil()]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: true }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_del_fil(), doc_del_fil(), doc_del_fil()]), @"Some((DocumentDeletion { deletion_ids: [0, 1, 2], includes_by_filter: true }, false))");
+    debug_snapshot!(autobatch_from(false,None,  [doc_del_fil()]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: true }, false))");
+    debug_snapshot!(autobatch_from(false,None,  [doc_del_fil(), doc_del_fil(), doc_del_fil()]), @"Some((DocumentDeletion { deletion_ids: [0, 1, 2], includes_by_filter: true }, false))");
+
+    // we can autobatch one or multiple Settings together
+    debug_snapshot!(autobatch_from(true, None, [settings(true)]), @"Some((Settings { allow_index_creation: true, settings_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [settings(true), settings(true), settings(true)]), @"Some((Settings { allow_index_creation: true, settings_ids: [0, 1, 2] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [settings(false)]), @"Some((Settings { allow_index_creation: false, settings_ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(true, None, [settings(false), settings(false), settings(false)]), @"Some((Settings { allow_index_creation: false, settings_ids: [0, 1, 2] }, false))");
+
+    debug_snapshot!(autobatch_from(false,None,  [settings(true)]), @"Some((Settings { allow_index_creation: true, settings_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(false,None,  [settings(true), settings(true), settings(true)]), @"Some((Settings { allow_index_creation: true, settings_ids: [0, 1, 2] }, true))");
+    debug_snapshot!(autobatch_from(false,None,  [settings(false)]), @"Some((Settings { allow_index_creation: false, settings_ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(false,None,  [settings(false), settings(false), settings(false)]), @"Some((Settings { allow_index_creation: false, settings_ids: [0, 1, 2] }, false))");
+
+    // We can autobatch document addition with document deletion
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None), doc_del()]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0, 1] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, true, None), doc_del()]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0, 1] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, false, None), doc_del()]), @"Some((DocumentOperation { allow_index_creation: false, primary_key: None, operation_ids: [0, 1] }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, false, None), doc_del()]), @"Some((DocumentOperation { allow_index_creation: false, primary_key: None, operation_ids: [0, 1] }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, Some("catto")), doc_del()]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("catto"), operation_ids: [0, 1] }, true))"###);
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, true, Some("catto")), doc_del()]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("catto"), operation_ids: [0, 1] }, true))"###);
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, false, Some("catto")), doc_del()]), @r###"Some((DocumentOperation { allow_index_creation: false, primary_key: Some("catto"), operation_ids: [0, 1] }, false))"###);
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, false, Some("catto")), doc_del()]), @r###"Some((DocumentOperation { allow_index_creation: false, primary_key: Some("catto"), operation_ids: [0, 1] }, false))"###);
+    debug_snapshot!(autobatch_from(false, None, [doc_imp(ReplaceDocuments, true, None), doc_del()]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0, 1] }, true))");
+    debug_snapshot!(autobatch_from(false, None, [doc_imp(UpdateDocuments, true, None), doc_del()]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0, 1] }, true))");
+    debug_snapshot!(autobatch_from(false, None, [doc_imp(ReplaceDocuments, false, None), doc_del()]), @"Some((DocumentOperation { allow_index_creation: false, primary_key: None, operation_ids: [0, 1] }, false))");
+    debug_snapshot!(autobatch_from(false, None, [doc_imp(UpdateDocuments, false, None), doc_del()]), @"Some((DocumentOperation { allow_index_creation: false, primary_key: None, operation_ids: [0, 1] }, false))");
+    debug_snapshot!(autobatch_from(false, None, [doc_imp(ReplaceDocuments, true, Some("catto")), doc_del()]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("catto"), operation_ids: [0, 1] }, true))"###);
+    debug_snapshot!(autobatch_from(false, None, [doc_imp(UpdateDocuments, true, Some("catto")), doc_del()]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("catto"), operation_ids: [0, 1] }, true))"###);
+    debug_snapshot!(autobatch_from(false, None, [doc_imp(ReplaceDocuments, false, Some("catto")), doc_del()]), @r###"Some((DocumentOperation { allow_index_creation: false, primary_key: Some("catto"), operation_ids: [0, 1] }, false))"###);
+    debug_snapshot!(autobatch_from(false, None, [doc_imp(UpdateDocuments, false, Some("catto")), doc_del()]), @r###"Some((DocumentOperation { allow_index_creation: false, primary_key: Some("catto"), operation_ids: [0, 1] }, false))"###);
+    // And the other way around
+    debug_snapshot!(autobatch_from(true, None, [doc_del(), doc_imp(ReplaceDocuments, true, None)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0, 1] }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_del(), doc_imp(UpdateDocuments, true, None)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0, 1] }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_del(), doc_imp(ReplaceDocuments, false, None)]), @"Some((DocumentOperation { allow_index_creation: false, primary_key: None, operation_ids: [0, 1] }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_del(), doc_imp(UpdateDocuments, false, None)]), @"Some((DocumentOperation { allow_index_creation: false, primary_key: None, operation_ids: [0, 1] }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_del(), doc_imp(ReplaceDocuments, true, Some("catto"))]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("catto"), operation_ids: [0, 1] }, false))"###);
+    debug_snapshot!(autobatch_from(true, None, [doc_del(), doc_imp(UpdateDocuments, true, Some("catto"))]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("catto"), operation_ids: [0, 1] }, false))"###);
+    debug_snapshot!(autobatch_from(true, None, [doc_del(), doc_imp(ReplaceDocuments, false, Some("catto"))]), @r###"Some((DocumentOperation { allow_index_creation: false, primary_key: Some("catto"), operation_ids: [0, 1] }, false))"###);
+    debug_snapshot!(autobatch_from(true, None, [doc_del(), doc_imp(UpdateDocuments, false, Some("catto"))]), @r###"Some((DocumentOperation { allow_index_creation: false, primary_key: Some("catto"), operation_ids: [0, 1] }, false))"###);
+    debug_snapshot!(autobatch_from(false, None, [doc_del(), doc_imp(ReplaceDocuments, false, None)]), @"Some((DocumentOperation { allow_index_creation: false, primary_key: None, operation_ids: [0, 1] }, false))");
+    debug_snapshot!(autobatch_from(false, None, [doc_del(), doc_imp(UpdateDocuments, false, None)]), @"Some((DocumentOperation { allow_index_creation: false, primary_key: None, operation_ids: [0, 1] }, false))");
+    debug_snapshot!(autobatch_from(false, None, [doc_del(), doc_imp(ReplaceDocuments, false, Some("catto"))]), @r###"Some((DocumentOperation { allow_index_creation: false, primary_key: Some("catto"), operation_ids: [0, 1] }, false))"###);
+    debug_snapshot!(autobatch_from(false, None, [doc_del(), doc_imp(UpdateDocuments, false, Some("catto"))]), @r###"Some((DocumentOperation { allow_index_creation: false, primary_key: Some("catto"), operation_ids: [0, 1] }, false))"###);
+
+    // But we can't autobatch document addition with document deletion by filter
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None), doc_del_fil()]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, true, None), doc_del_fil()]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, false, None), doc_del_fil()]), @"Some((DocumentOperation { allow_index_creation: false, primary_key: None, operation_ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, false, None), doc_del_fil()]), @"Some((DocumentOperation { allow_index_creation: false, primary_key: None, operation_ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, Some("catto")), doc_del_fil()]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("catto"), operation_ids: [0] }, true))"###);
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, true, Some("catto")), doc_del_fil()]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("catto"), operation_ids: [0] }, true))"###);
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, false, Some("catto")), doc_del_fil()]), @r###"Some((DocumentOperation { allow_index_creation: false, primary_key: Some("catto"), operation_ids: [0] }, false))"###);
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, false, Some("catto")), doc_del_fil()]), @r###"Some((DocumentOperation { allow_index_creation: false, primary_key: Some("catto"), operation_ids: [0] }, false))"###);
+    debug_snapshot!(autobatch_from(false, None, [doc_imp(ReplaceDocuments, true, None), doc_del_fil()]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(false, None, [doc_imp(UpdateDocuments, true, None), doc_del_fil()]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(false, None, [doc_imp(ReplaceDocuments, false, None), doc_del_fil()]), @"Some((DocumentOperation { allow_index_creation: false, primary_key: None, operation_ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(false, None, [doc_imp(UpdateDocuments, false, None), doc_del_fil()]), @"Some((DocumentOperation { allow_index_creation: false, primary_key: None, operation_ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(false, None, [doc_imp(ReplaceDocuments, true, Some("catto")), doc_del_fil()]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("catto"), operation_ids: [0] }, true))"###);
+    debug_snapshot!(autobatch_from(false, None, [doc_imp(UpdateDocuments, true, Some("catto")), doc_del_fil()]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("catto"), operation_ids: [0] }, true))"###);
+    debug_snapshot!(autobatch_from(false, None, [doc_imp(ReplaceDocuments, false, Some("catto")), doc_del_fil()]), @r###"Some((DocumentOperation { allow_index_creation: false, primary_key: Some("catto"), operation_ids: [0] }, false))"###);
+    debug_snapshot!(autobatch_from(false, None, [doc_imp(UpdateDocuments, false, Some("catto")), doc_del_fil()]), @r###"Some((DocumentOperation { allow_index_creation: false, primary_key: Some("catto"), operation_ids: [0] }, false))"###);
+    // And the other way around
+    debug_snapshot!(autobatch_from(true, None, [doc_del_fil(), doc_imp(ReplaceDocuments, true, None)]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: true }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_del_fil(), doc_imp(UpdateDocuments, true, None)]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: true }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_del_fil(), doc_imp(ReplaceDocuments, false, None)]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: true }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_del_fil(), doc_imp(UpdateDocuments, false, None)]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: true }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_del_fil(), doc_imp(ReplaceDocuments, true, Some("catto"))]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: true }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_del_fil(), doc_imp(UpdateDocuments, true, Some("catto"))]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: true }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_del_fil(), doc_imp(ReplaceDocuments, false, Some("catto"))]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: true }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_del_fil(), doc_imp(UpdateDocuments, false, Some("catto"))]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: true }, false))");
+    debug_snapshot!(autobatch_from(false, None, [doc_del_fil(), doc_imp(ReplaceDocuments, false, None)]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: true }, false))");
+    debug_snapshot!(autobatch_from(false, None, [doc_del_fil(), doc_imp(UpdateDocuments, false, None)]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: true }, false))");
+    debug_snapshot!(autobatch_from(false, None, [doc_del_fil(), doc_imp(ReplaceDocuments, false, Some("catto"))]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: true }, false))");
+    debug_snapshot!(autobatch_from(false, None, [doc_del_fil(), doc_imp(UpdateDocuments, false, Some("catto"))]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: true }, false))");
+}
+
+#[test]
+fn simple_different_document_operations_autobatch_together() {
+    // addition and updates with deletion by filter can't batch together
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None), doc_imp(UpdateDocuments, true, None)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0, 1] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, true, None), doc_imp(ReplaceDocuments, true, None)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0, 1] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, true, None), doc_del_fil()]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None), doc_del_fil()]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_del_fil(), doc_imp(UpdateDocuments, true, None)]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: true }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_del_fil(), doc_imp(ReplaceDocuments, true, None)]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: true }, false))");
+
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None), idx_create()]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, true, None), idx_create()]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_del(), idx_create()]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: false }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_del_fil(), idx_create()]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: true }, false))");
+
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None), idx_update()]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, true, None), idx_update()]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_del(), idx_update()]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: false }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_del_fil(), idx_update()]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: true }, false))");
+
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None), idx_swap()]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, true, None), idx_swap()]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_del(), idx_swap()]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: false }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_del_fil(), idx_swap()]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: true }, false))");
+}
+
+#[test]
+fn document_addition_doesnt_batch_with_settings() {
+    // simple case
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None), settings(true)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, true, None), settings(true)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+
+    // multiple settings and doc addition
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None), doc_imp(ReplaceDocuments, true, None), settings(true), settings(true)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0, 1] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None), doc_imp(ReplaceDocuments, true, None), settings(true), settings(true)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0, 1] }, true))");
+
+    // addition and setting unordered
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None), settings(true), doc_imp(ReplaceDocuments, true, None), settings(true)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, true, None), settings(true), doc_imp(UpdateDocuments, true, None), settings(true)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+
+    // Doesn't batch with other forbidden operations
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None), settings(true), doc_imp(UpdateDocuments, true, None)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, true, None), settings(true), doc_imp(ReplaceDocuments, true, None)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None), settings(true), doc_del()]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, true, None), settings(true), doc_del()]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None), settings(true), idx_create()]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, true, None), settings(true), idx_create()]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None), settings(true), idx_update()]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, true, None), settings(true), idx_update()]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None), settings(true), idx_swap()]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, true, None), settings(true), idx_swap()]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+}
+
+#[test]
+fn clear_and_additions() {
+    // these two doesn't need to batch
+    debug_snapshot!(autobatch_from(true, None, [doc_clr(), doc_imp(ReplaceDocuments, true, None)]), @"Some((DocumentClear { ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_clr(), doc_imp(UpdateDocuments, true, None)]), @"Some((DocumentClear { ids: [0] }, false))");
+
+    // Basic use case
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None), doc_imp(ReplaceDocuments, true, None), doc_clr()]), @"Some((DocumentClear { ids: [0, 1, 2] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, true, None), doc_imp(UpdateDocuments, true, None), doc_clr()]), @"Some((DocumentClear { ids: [0, 1, 2] }, true))");
+
+    // This batch kind doesn't mix with other document addition
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None), doc_imp(ReplaceDocuments, true, None), doc_clr(), doc_imp(ReplaceDocuments, true, None)]), @"Some((DocumentClear { ids: [0, 1, 2] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, true, None), doc_imp(UpdateDocuments, true, None), doc_clr(), doc_imp(UpdateDocuments, true, None)]), @"Some((DocumentClear { ids: [0, 1, 2] }, true))");
+
+    // But you can batch multiple clear together
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None), doc_imp(ReplaceDocuments, true, None), doc_clr(), doc_clr(), doc_clr()]), @"Some((DocumentClear { ids: [0, 1, 2, 3, 4] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, true, None), doc_imp(UpdateDocuments, true, None), doc_clr(), doc_clr(), doc_clr()]), @"Some((DocumentClear { ids: [0, 1, 2, 3, 4] }, true))");
+}
+
+#[test]
+fn clear_and_additions_and_settings() {
+    // A clear don't need to autobatch the settings that happens AFTER there is no documents
+    debug_snapshot!(autobatch_from(true, None, [doc_clr(), settings(true)]), @"Some((DocumentClear { ids: [0] }, false))");
+
+    debug_snapshot!(autobatch_from(true, None, [settings(true), doc_clr(), settings(true)]), @"Some((ClearAndSettings { other: [1], allow_index_creation: true, settings_ids: [0, 2] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None), settings(true), doc_clr()]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, true, None), settings(true), doc_clr()]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+}
+
+#[test]
+fn anything_and_index_deletion() {
+    // The `IndexDeletion` doesn't batch with anything that happens AFTER.
+    debug_snapshot!(autobatch_from(true, None, [idx_del(), doc_imp(ReplaceDocuments, true, None)]), @"Some((IndexDeletion { ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(true, None, [idx_del(), doc_imp(UpdateDocuments, true, None)]), @"Some((IndexDeletion { ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(true, None, [idx_del(), doc_imp(ReplaceDocuments, false, None)]), @"Some((IndexDeletion { ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(true, None, [idx_del(), doc_imp(UpdateDocuments, false, None)]), @"Some((IndexDeletion { ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(true, None, [idx_del(), doc_del()]), @"Some((IndexDeletion { ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(true, None, [idx_del(), doc_del_fil()]), @"Some((IndexDeletion { ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(true, None, [idx_del(), doc_clr()]), @"Some((IndexDeletion { ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(true, None, [idx_del(), settings(true)]), @"Some((IndexDeletion { ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(true, None, [idx_del(), settings(false)]), @"Some((IndexDeletion { ids: [0] }, false))");
+
+    debug_snapshot!(autobatch_from(false,None,  [idx_del(), doc_imp(ReplaceDocuments, true, None)]), @"Some((IndexDeletion { ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(false,None,  [idx_del(), doc_imp(UpdateDocuments, true, None)]), @"Some((IndexDeletion { ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(false,None,  [idx_del(), doc_imp(ReplaceDocuments, false, None)]), @"Some((IndexDeletion { ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(false,None,  [idx_del(), doc_imp(UpdateDocuments, false, None)]), @"Some((IndexDeletion { ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(false,None,  [idx_del(), doc_del()]), @"Some((IndexDeletion { ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(false,None,  [idx_del(), doc_del_fil()]), @"Some((IndexDeletion { ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(false,None,  [idx_del(), doc_clr()]), @"Some((IndexDeletion { ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(false,None,  [idx_del(), settings(true)]), @"Some((IndexDeletion { ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(false,None,  [idx_del(), settings(false)]), @"Some((IndexDeletion { ids: [0] }, false))");
+
+    // The index deletion can accept almost any type of `BatchKind` and transform it to an `IndexDeletion`.
+    // First, the basic cases
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None), idx_del()]), @"Some((IndexDeletion { ids: [0, 1] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, true, None), idx_del()]), @"Some((IndexDeletion { ids: [0, 1] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, false, None), idx_del()]), @"Some((IndexDeletion { ids: [0, 1] }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(UpdateDocuments, false, None), idx_del()]), @"Some((IndexDeletion { ids: [0, 1] }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_del(), idx_del()]), @"Some((IndexDeletion { ids: [0, 1] }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_del_fil(), idx_del()]), @"Some((IndexDeletion { ids: [0, 1] }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_clr(), idx_del()]), @"Some((IndexDeletion { ids: [0, 1] }, false))");
+    debug_snapshot!(autobatch_from(true, None, [settings(true), idx_del()]), @"Some((IndexDeletion { ids: [0, 1] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [settings(false), idx_del()]), @"Some((IndexDeletion { ids: [0, 1] }, false))");
+
+    debug_snapshot!(autobatch_from(false,None,  [doc_imp(ReplaceDocuments, true, None), idx_del()]), @"Some((IndexDeletion { ids: [0, 1] }, true))");
+    debug_snapshot!(autobatch_from(false,None,  [doc_imp(UpdateDocuments, true, None), idx_del()]), @"Some((IndexDeletion { ids: [0, 1] }, true))");
+    debug_snapshot!(autobatch_from(false,None,  [doc_imp(ReplaceDocuments, false, None), idx_del()]), @"Some((IndexDeletion { ids: [0, 1] }, false))");
+    debug_snapshot!(autobatch_from(false,None,  [doc_imp(UpdateDocuments, false, None), idx_del()]), @"Some((IndexDeletion { ids: [0, 1] }, false))");
+    debug_snapshot!(autobatch_from(false,None,  [doc_del(), idx_del()]), @"Some((IndexDeletion { ids: [0, 1] }, false))");
+    debug_snapshot!(autobatch_from(false,None,  [doc_del_fil(), idx_del()]), @"Some((IndexDeletion { ids: [0, 1] }, false))");
+    debug_snapshot!(autobatch_from(false,None,  [doc_clr(), idx_del()]), @"Some((IndexDeletion { ids: [0, 1] }, false))");
+    debug_snapshot!(autobatch_from(false,None,  [settings(true), idx_del()]), @"Some((IndexDeletion { ids: [0, 1] }, true))");
+    debug_snapshot!(autobatch_from(false,None,  [settings(false), idx_del()]), @"Some((IndexDeletion { ids: [0, 1] }, false))");
+}
+
+#[test]
+fn allowed_and_disallowed_index_creation() {
+    // `DocumentImport` can't be mixed with those disallowed to do so except if the index already exists.
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, false, None), doc_imp(ReplaceDocuments, true, None)]), @"Some((DocumentOperation { allow_index_creation: false, primary_key: None, operation_ids: [0, 1] }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None), doc_imp(ReplaceDocuments, true, None)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0, 1] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, false, None), doc_imp(ReplaceDocuments, false, None)]), @"Some((DocumentOperation { allow_index_creation: false, primary_key: None, operation_ids: [0, 1] }, false))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None), settings(true)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, false, None), settings(true)]), @"Some((DocumentOperation { allow_index_creation: false, primary_key: None, operation_ids: [0] }, false))");
+
+    debug_snapshot!(autobatch_from(false,None,  [doc_imp(ReplaceDocuments, false, None), doc_imp(ReplaceDocuments, true, None)]), @"Some((DocumentOperation { allow_index_creation: false, primary_key: None, operation_ids: [0] }, false))");
+    debug_snapshot!(autobatch_from(false,None,  [doc_imp(ReplaceDocuments, true, None), doc_imp(ReplaceDocuments, true, None)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0, 1] }, true))");
+    debug_snapshot!(autobatch_from(false,None,  [doc_imp(ReplaceDocuments, false, None), doc_imp(ReplaceDocuments, false, None)]), @"Some((DocumentOperation { allow_index_creation: false, primary_key: None, operation_ids: [0, 1] }, false))");
+    debug_snapshot!(autobatch_from(false,None,  [doc_imp(ReplaceDocuments, true, None), settings(true)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(false,None,  [doc_imp(ReplaceDocuments, false, None), settings(true)]), @"Some((DocumentOperation { allow_index_creation: false, primary_key: None, operation_ids: [0] }, false))");
+
+    // batch deletion and addition
+    debug_snapshot!(autobatch_from(false, None, [doc_del(), doc_imp(ReplaceDocuments, true, Some("catto"))]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: false }, false))");
+    debug_snapshot!(autobatch_from(false, None, [doc_del(), doc_imp(UpdateDocuments, true, Some("catto"))]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: false }, false))");
+    debug_snapshot!(autobatch_from(false, None, [doc_del(), doc_imp(ReplaceDocuments, true, None)]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: false }, false))");
+    debug_snapshot!(autobatch_from(false, None, [doc_del(), doc_imp(UpdateDocuments, true, None)]), @"Some((DocumentDeletion { deletion_ids: [0], includes_by_filter: false }, false))");
+}
+
+#[test]
+fn autobatch_primary_key() {
+    // ==> If I have a pk
+    // With a single update
+    debug_snapshot!(autobatch_from(true, Some("id"), [doc_imp(ReplaceDocuments, true, None)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, Some("id"), [doc_imp(ReplaceDocuments, true, Some("id"))]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("id"), operation_ids: [0] }, true))"###);
+    debug_snapshot!(autobatch_from(true, Some("id"), [doc_imp(ReplaceDocuments, true, Some("other"))]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("other"), operation_ids: [0] }, true))"###);
+
+    // With a multiple updates
+    debug_snapshot!(autobatch_from(true, Some("id"), [doc_imp(ReplaceDocuments, true, None), doc_imp(ReplaceDocuments, true, None)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0, 1] }, true))");
+    debug_snapshot!(autobatch_from(true, Some("id"), [doc_imp(ReplaceDocuments, true, None), doc_imp(ReplaceDocuments, true, Some("id"))]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("id"), operation_ids: [0, 1] }, true))"###);
+    debug_snapshot!(autobatch_from(true, Some("id"), [doc_imp(ReplaceDocuments, true, None), doc_imp(ReplaceDocuments, true, Some("id")), doc_imp(ReplaceDocuments, true, None)]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("id"), operation_ids: [0, 1] }, true))"###);
+    debug_snapshot!(autobatch_from(true, Some("id"), [doc_imp(ReplaceDocuments, true, None), doc_imp(ReplaceDocuments, true, Some("other"))]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, Some("id"), [doc_imp(ReplaceDocuments, true, None), doc_imp(ReplaceDocuments, true, Some("other")), doc_imp(ReplaceDocuments, true, None)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, Some("id"), [doc_imp(ReplaceDocuments, true, None), doc_imp(ReplaceDocuments, true, Some("other")), doc_imp(ReplaceDocuments, true, Some("id"))]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+
+    debug_snapshot!(autobatch_from(true, Some("id"), [doc_imp(ReplaceDocuments, true, Some("id")), doc_imp(ReplaceDocuments, true, None)]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("id"), operation_ids: [0] }, true))"###);
+    debug_snapshot!(autobatch_from(true, Some("id"), [doc_imp(ReplaceDocuments, true, Some("id")), doc_imp(ReplaceDocuments, true, Some("id"))]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("id"), operation_ids: [0, 1] }, true))"###);
+    debug_snapshot!(autobatch_from(true, Some("id"), [doc_imp(ReplaceDocuments, true, Some("id")), doc_imp(ReplaceDocuments, true, Some("id")), doc_imp(ReplaceDocuments, true, None)]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("id"), operation_ids: [0, 1] }, true))"###);
+    debug_snapshot!(autobatch_from(true, Some("id"), [doc_imp(ReplaceDocuments, true, Some("id")), doc_imp(ReplaceDocuments, true, Some("other"))]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("id"), operation_ids: [0] }, true))"###);
+    debug_snapshot!(autobatch_from(true, Some("id"), [doc_imp(ReplaceDocuments, true, Some("id")), doc_imp(ReplaceDocuments, true, Some("other")), doc_imp(ReplaceDocuments, true, None)]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("id"), operation_ids: [0] }, true))"###);
+    debug_snapshot!(autobatch_from(true, Some("id"), [doc_imp(ReplaceDocuments, true, Some("id")), doc_imp(ReplaceDocuments, true, Some("other")), doc_imp(ReplaceDocuments, true, Some("id"))]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("id"), operation_ids: [0] }, true))"###);
+
+    debug_snapshot!(autobatch_from(true, Some("id"), [doc_imp(ReplaceDocuments, true, Some("other")), doc_imp(ReplaceDocuments, true, None)]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("other"), operation_ids: [0] }, true))"###);
+    debug_snapshot!(autobatch_from(true, Some("id"), [doc_imp(ReplaceDocuments, true, Some("other")), doc_imp(ReplaceDocuments, true, Some("id"))]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("other"), operation_ids: [0] }, true))"###);
+    debug_snapshot!(autobatch_from(true, Some("id"), [doc_imp(ReplaceDocuments, true, Some("other")), doc_imp(ReplaceDocuments, true, Some("id")), doc_imp(ReplaceDocuments, true, None)]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("other"), operation_ids: [0] }, true))"###);
+    debug_snapshot!(autobatch_from(true, Some("id"), [doc_imp(ReplaceDocuments, true, Some("other")), doc_imp(ReplaceDocuments, true, Some("other"))]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("other"), operation_ids: [0] }, true))"###);
+    debug_snapshot!(autobatch_from(true, Some("id"), [doc_imp(ReplaceDocuments, true, Some("other")), doc_imp(ReplaceDocuments, true, Some("other")), doc_imp(ReplaceDocuments, true, None)]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("other"), operation_ids: [0] }, true))"###);
+    debug_snapshot!(autobatch_from(true, Some("id"), [doc_imp(ReplaceDocuments, true, Some("other")), doc_imp(ReplaceDocuments, true, Some("other")), doc_imp(ReplaceDocuments, true, Some("id"))]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("other"), operation_ids: [0] }, true))"###);
+
+    // ==> If I don't have a pk
+    // With a single update
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, Some("id"))]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("id"), operation_ids: [0] }, true))"###);
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, Some("other"))]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("other"), operation_ids: [0] }, true))"###);
+
+    // With a multiple updates
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None), doc_imp(ReplaceDocuments, true, None)]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0, 1] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, None), doc_imp(ReplaceDocuments, true, Some("id"))]), @"Some((DocumentOperation { allow_index_creation: true, primary_key: None, operation_ids: [0] }, true))");
+    debug_snapshot!(autobatch_from(true, None, [doc_imp(ReplaceDocuments, true, Some("id")), doc_imp(ReplaceDocuments, true, None)]), @r###"Some((DocumentOperation { allow_index_creation: true, primary_key: Some("id"), operation_ids: [0] }, true))"###);
+}

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

@@ -0,0 +1,567 @@
+use std::fmt;
+
+use meilisearch_types::heed::RoTxn;
+use meilisearch_types::milli::update::IndexDocumentsMethod;
+use meilisearch_types::settings::{Settings, Unchecked};
+use meilisearch_types::tasks::{Kind, KindWithContent, Status, Task};
+use roaring::RoaringBitmap;
+use uuid::Uuid;
+
+use super::autobatcher::{self, BatchKind};
+use crate::utils::ProcessingBatch;
+use crate::{Error, IndexScheduler, Result};
+
+/// Represents a combination of tasks that can all be processed at the same time.
+///
+/// A batch contains the set of tasks that it represents (accessible through
+/// [`self.ids()`](Batch::ids)), as well as additional information on how to
+/// be processed.
+#[derive(Debug)]
+pub(crate) enum Batch {
+    TaskCancelation {
+        /// The task cancelation itself.
+        task: Task,
+    },
+    TaskDeletions(Vec<Task>),
+    SnapshotCreation(Vec<Task>),
+    Dump(Task),
+    IndexOperation {
+        op: IndexOperation,
+        must_create_index: bool,
+    },
+    IndexCreation {
+        index_uid: String,
+        primary_key: Option<String>,
+        task: Task,
+    },
+    IndexUpdate {
+        index_uid: String,
+        primary_key: Option<String>,
+        task: Task,
+    },
+    IndexDeletion {
+        index_uid: String,
+        tasks: Vec<Task>,
+        index_has_been_created: bool,
+    },
+    IndexSwap {
+        task: Task,
+    },
+    UpgradeDatabase {
+        tasks: Vec<Task>,
+    },
+}
+
+#[derive(Debug)]
+pub(crate) enum DocumentOperation {
+    Replace(Uuid),
+    Update(Uuid),
+    Delete(Vec<String>),
+}
+
+/// A [batch](Batch) that combines multiple tasks operating on an index.
+#[derive(Debug)]
+pub(crate) enum IndexOperation {
+    DocumentOperation {
+        index_uid: String,
+        primary_key: Option<String>,
+        operations: Vec<DocumentOperation>,
+        tasks: Vec<Task>,
+    },
+    DocumentEdition {
+        index_uid: String,
+        task: Task,
+    },
+    DocumentDeletion {
+        index_uid: String,
+        tasks: Vec<Task>,
+    },
+    DocumentClear {
+        index_uid: String,
+        tasks: Vec<Task>,
+    },
+    Settings {
+        index_uid: String,
+        // The boolean indicates if it's a settings deletion or creation.
+        settings: Vec<(bool, Settings<Unchecked>)>,
+        tasks: Vec<Task>,
+    },
+    DocumentClearAndSetting {
+        index_uid: String,
+        cleared_tasks: Vec<Task>,
+
+        // The boolean indicates if it's a settings deletion or creation.
+        settings: Vec<(bool, Settings<Unchecked>)>,
+        settings_tasks: Vec<Task>,
+    },
+}
+
+impl Batch {
+    /// Return the task ids associated with this batch.
+    pub fn ids(&self) -> RoaringBitmap {
+        match self {
+            Batch::TaskCancelation { task, .. }
+            | Batch::Dump(task)
+            | Batch::IndexCreation { task, .. }
+            | Batch::IndexUpdate { task, .. } => {
+                RoaringBitmap::from_sorted_iter(std::iter::once(task.uid)).unwrap()
+            }
+            Batch::SnapshotCreation(tasks)
+            | Batch::TaskDeletions(tasks)
+            | Batch::UpgradeDatabase { tasks }
+            | Batch::IndexDeletion { tasks, .. } => {
+                RoaringBitmap::from_iter(tasks.iter().map(|task| task.uid))
+            }
+            Batch::IndexOperation { op, .. } => match op {
+                IndexOperation::DocumentOperation { tasks, .. }
+                | IndexOperation::Settings { tasks, .. }
+                | IndexOperation::DocumentDeletion { tasks, .. }
+                | IndexOperation::DocumentClear { tasks, .. } => {
+                    RoaringBitmap::from_iter(tasks.iter().map(|task| task.uid))
+                }
+                IndexOperation::DocumentEdition { task, .. } => {
+                    RoaringBitmap::from_sorted_iter(std::iter::once(task.uid)).unwrap()
+                }
+                IndexOperation::DocumentClearAndSetting {
+                    cleared_tasks: tasks,
+                    settings_tasks: other,
+                    ..
+                } => RoaringBitmap::from_iter(tasks.iter().chain(other).map(|task| task.uid)),
+            },
+            Batch::IndexSwap { task } => {
+                RoaringBitmap::from_sorted_iter(std::iter::once(task.uid)).unwrap()
+            }
+        }
+    }
+
+    /// Return the index UID associated with this batch
+    pub fn index_uid(&self) -> Option<&str> {
+        use Batch::*;
+        match self {
+            TaskCancelation { .. }
+            | TaskDeletions(_)
+            | SnapshotCreation(_)
+            | Dump(_)
+            | UpgradeDatabase { .. }
+            | IndexSwap { .. } => None,
+            IndexOperation { op, .. } => Some(op.index_uid()),
+            IndexCreation { index_uid, .. }
+            | IndexUpdate { index_uid, .. }
+            | IndexDeletion { index_uid, .. } => Some(index_uid),
+        }
+    }
+}
+
+impl fmt::Display for Batch {
+    /// A text used when we debug the profiling reports.
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        let index_uid = self.index_uid();
+        let tasks = self.ids();
+        match self {
+            Batch::TaskCancelation { .. } => f.write_str("TaskCancelation")?,
+            Batch::TaskDeletions(_) => f.write_str("TaskDeletion")?,
+            Batch::SnapshotCreation(_) => f.write_str("SnapshotCreation")?,
+            Batch::Dump(_) => f.write_str("Dump")?,
+            Batch::IndexOperation { op, .. } => write!(f, "{op}")?,
+            Batch::IndexCreation { .. } => f.write_str("IndexCreation")?,
+            Batch::IndexUpdate { .. } => f.write_str("IndexUpdate")?,
+            Batch::IndexDeletion { .. } => f.write_str("IndexDeletion")?,
+            Batch::IndexSwap { .. } => f.write_str("IndexSwap")?,
+            Batch::UpgradeDatabase { .. } => f.write_str("UpgradeDatabase")?,
+        };
+        match index_uid {
+            Some(name) => f.write_fmt(format_args!(" on {name:?} from tasks: {tasks:?}")),
+            None => f.write_fmt(format_args!(" from tasks: {tasks:?}")),
+        }
+    }
+}
+
+impl IndexOperation {
+    pub fn index_uid(&self) -> &str {
+        match self {
+            IndexOperation::DocumentOperation { index_uid, .. }
+            | IndexOperation::DocumentEdition { index_uid, .. }
+            | IndexOperation::DocumentDeletion { index_uid, .. }
+            | IndexOperation::DocumentClear { index_uid, .. }
+            | IndexOperation::Settings { index_uid, .. }
+            | IndexOperation::DocumentClearAndSetting { index_uid, .. } => index_uid,
+        }
+    }
+}
+
+impl fmt::Display for IndexOperation {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        match self {
+            IndexOperation::DocumentOperation { .. } => {
+                f.write_str("IndexOperation::DocumentOperation")
+            }
+            IndexOperation::DocumentEdition { .. } => {
+                f.write_str("IndexOperation::DocumentEdition")
+            }
+            IndexOperation::DocumentDeletion { .. } => {
+                f.write_str("IndexOperation::DocumentDeletion")
+            }
+            IndexOperation::DocumentClear { .. } => f.write_str("IndexOperation::DocumentClear"),
+            IndexOperation::Settings { .. } => f.write_str("IndexOperation::Settings"),
+            IndexOperation::DocumentClearAndSetting { .. } => {
+                f.write_str("IndexOperation::DocumentClearAndSetting")
+            }
+        }
+    }
+}
+
+impl IndexScheduler {
+    /// Convert an [`BatchKind`](crate::autobatcher::BatchKind) into a [`Batch`].
+    ///
+    /// ## Arguments
+    /// - `rtxn`: read transaction
+    /// - `index_uid`: name of the index affected by the operations of the autobatch
+    /// - `batch`: the result of the autobatcher
+    pub(crate) fn create_next_batch_index(
+        &self,
+        rtxn: &RoTxn,
+        index_uid: String,
+        batch: BatchKind,
+        current_batch: &mut ProcessingBatch,
+        must_create_index: bool,
+    ) -> Result<Option<Batch>> {
+        match batch {
+            BatchKind::DocumentClear { ids } => Ok(Some(Batch::IndexOperation {
+                op: IndexOperation::DocumentClear {
+                    tasks: self.queue.get_existing_tasks_for_processing_batch(
+                        rtxn,
+                        current_batch,
+                        ids,
+                    )?,
+                    index_uid,
+                },
+                must_create_index,
+            })),
+            BatchKind::DocumentEdition { id } => {
+                let mut task =
+                    self.queue.tasks.get_task(rtxn, id)?.ok_or(Error::CorruptedTaskQueue)?;
+                current_batch.processing(Some(&mut task));
+                match &task.kind {
+                    KindWithContent::DocumentEdition { index_uid, .. } => {
+                        Ok(Some(Batch::IndexOperation {
+                            op: IndexOperation::DocumentEdition {
+                                index_uid: index_uid.clone(),
+                                task,
+                            },
+                            must_create_index: false,
+                        }))
+                    }
+                    _ => unreachable!(),
+                }
+            }
+            BatchKind::DocumentOperation { operation_ids, .. } => {
+                let tasks = self.queue.get_existing_tasks_for_processing_batch(
+                    rtxn,
+                    current_batch,
+                    operation_ids,
+                )?;
+                let primary_key = tasks
+                    .iter()
+                    .find_map(|task| match task.kind {
+                        KindWithContent::DocumentAdditionOrUpdate { ref primary_key, .. } => {
+                            // we want to stop on the first document addition
+                            Some(primary_key.clone())
+                        }
+                        KindWithContent::DocumentDeletion { .. } => None,
+                        _ => unreachable!(),
+                    })
+                    .flatten();
+
+                let mut operations = Vec::new();
+
+                for task in tasks.iter() {
+                    match task.kind {
+                        KindWithContent::DocumentAdditionOrUpdate {
+                            content_file, method, ..
+                        } => match method {
+                            IndexDocumentsMethod::ReplaceDocuments => {
+                                operations.push(DocumentOperation::Replace(content_file))
+                            }
+                            IndexDocumentsMethod::UpdateDocuments => {
+                                operations.push(DocumentOperation::Update(content_file))
+                            }
+                            _ => unreachable!("Unknown document merging method"),
+                        },
+                        KindWithContent::DocumentDeletion { ref documents_ids, .. } => {
+                            operations.push(DocumentOperation::Delete(documents_ids.clone()));
+                        }
+                        _ => unreachable!(),
+                    }
+                }
+
+                Ok(Some(Batch::IndexOperation {
+                    op: IndexOperation::DocumentOperation {
+                        index_uid,
+                        primary_key,
+                        operations,
+                        tasks,
+                    },
+                    must_create_index,
+                }))
+            }
+            BatchKind::DocumentDeletion { deletion_ids, includes_by_filter: _ } => {
+                let tasks = self.queue.get_existing_tasks_for_processing_batch(
+                    rtxn,
+                    current_batch,
+                    deletion_ids,
+                )?;
+
+                Ok(Some(Batch::IndexOperation {
+                    op: IndexOperation::DocumentDeletion { index_uid, tasks },
+                    must_create_index,
+                }))
+            }
+            BatchKind::Settings { settings_ids, .. } => {
+                let tasks = self.queue.get_existing_tasks_for_processing_batch(
+                    rtxn,
+                    current_batch,
+                    settings_ids,
+                )?;
+
+                let mut settings = Vec::new();
+                for task in &tasks {
+                    match task.kind {
+                        KindWithContent::SettingsUpdate {
+                            ref new_settings, is_deletion, ..
+                        } => settings.push((is_deletion, *new_settings.clone())),
+                        _ => unreachable!(),
+                    }
+                }
+
+                Ok(Some(Batch::IndexOperation {
+                    op: IndexOperation::Settings { index_uid, settings, tasks },
+                    must_create_index,
+                }))
+            }
+            BatchKind::ClearAndSettings { other, settings_ids, allow_index_creation } => {
+                let (index_uid, settings, settings_tasks) = match self
+                    .create_next_batch_index(
+                        rtxn,
+                        index_uid,
+                        BatchKind::Settings { settings_ids, allow_index_creation },
+                        current_batch,
+                        must_create_index,
+                    )?
+                    .unwrap()
+                {
+                    Batch::IndexOperation {
+                        op: IndexOperation::Settings { index_uid, settings, tasks, .. },
+                        ..
+                    } => (index_uid, settings, tasks),
+                    _ => unreachable!(),
+                };
+                let (index_uid, cleared_tasks) = match self
+                    .create_next_batch_index(
+                        rtxn,
+                        index_uid,
+                        BatchKind::DocumentClear { ids: other },
+                        current_batch,
+                        must_create_index,
+                    )?
+                    .unwrap()
+                {
+                    Batch::IndexOperation {
+                        op: IndexOperation::DocumentClear { index_uid, tasks },
+                        ..
+                    } => (index_uid, tasks),
+                    _ => unreachable!(),
+                };
+
+                Ok(Some(Batch::IndexOperation {
+                    op: IndexOperation::DocumentClearAndSetting {
+                        index_uid,
+                        cleared_tasks,
+                        settings,
+                        settings_tasks,
+                    },
+                    must_create_index,
+                }))
+            }
+            BatchKind::IndexCreation { id } => {
+                let mut task =
+                    self.queue.tasks.get_task(rtxn, id)?.ok_or(Error::CorruptedTaskQueue)?;
+                current_batch.processing(Some(&mut task));
+                let (index_uid, primary_key) = match &task.kind {
+                    KindWithContent::IndexCreation { index_uid, primary_key } => {
+                        (index_uid.clone(), primary_key.clone())
+                    }
+                    _ => unreachable!(),
+                };
+                Ok(Some(Batch::IndexCreation { index_uid, primary_key, task }))
+            }
+            BatchKind::IndexUpdate { id } => {
+                let mut task =
+                    self.queue.tasks.get_task(rtxn, id)?.ok_or(Error::CorruptedTaskQueue)?;
+                current_batch.processing(Some(&mut task));
+                let primary_key = match &task.kind {
+                    KindWithContent::IndexUpdate { primary_key, .. } => primary_key.clone(),
+                    _ => unreachable!(),
+                };
+                Ok(Some(Batch::IndexUpdate { index_uid, primary_key, task }))
+            }
+            BatchKind::IndexDeletion { ids } => Ok(Some(Batch::IndexDeletion {
+                index_uid,
+                index_has_been_created: must_create_index,
+                tasks: self.queue.get_existing_tasks_for_processing_batch(
+                    rtxn,
+                    current_batch,
+                    ids,
+                )?,
+            })),
+            BatchKind::IndexSwap { id } => {
+                let mut task =
+                    self.queue.tasks.get_task(rtxn, id)?.ok_or(Error::CorruptedTaskQueue)?;
+                current_batch.processing(Some(&mut task));
+                Ok(Some(Batch::IndexSwap { task }))
+            }
+        }
+    }
+
+    /// Create the next batch to be processed;
+    /// 1. We get the *last* task to cancel.
+    /// 2. We get the *next* task to delete.
+    /// 3. We get the *next* snapshot to process.
+    /// 4. We get the *next* dump to process.
+    /// 5. We get the *next* tasks to process for a specific index.
+    #[tracing::instrument(level = "trace", skip(self, rtxn), target = "indexing::scheduler")]
+    pub(crate) fn create_next_batch(
+        &self,
+        rtxn: &RoTxn,
+    ) -> Result<Option<(Batch, ProcessingBatch)>> {
+        #[cfg(test)]
+        self.maybe_fail(crate::test_utils::FailureLocation::InsideCreateBatch)?;
+
+        let batch_id = self.queue.batches.next_batch_id(rtxn)?;
+        let mut current_batch = ProcessingBatch::new(batch_id);
+
+        let enqueued = &self.queue.tasks.get_status(rtxn, Status::Enqueued)?;
+        let failed = &self.queue.tasks.get_status(rtxn, Status::Failed)?;
+
+        // 0. The priority over everything is to upgrade the instance
+        // There shouldn't be multiple upgrade tasks but just in case we're going to batch all of them at the same time
+        let upgrade = self.queue.tasks.get_kind(rtxn, Kind::UpgradeDatabase)? & (enqueued | failed);
+        if !upgrade.is_empty() {
+            let mut tasks = self.queue.tasks.get_existing_tasks(rtxn, upgrade)?;
+            // In the case of an upgrade database batch, we want to find back the original batch that tried processing it
+            // and re-use its id
+            if let Some(batch_uid) = tasks.last().unwrap().batch_uid {
+                current_batch.uid = batch_uid;
+            }
+            current_batch.processing(&mut tasks);
+            return Ok(Some((Batch::UpgradeDatabase { tasks }, current_batch)));
+        }
+
+        // 1. we get the last task to cancel.
+        let to_cancel = self.queue.tasks.get_kind(rtxn, Kind::TaskCancelation)? & enqueued;
+        if let Some(task_id) = to_cancel.max() {
+            let mut task =
+                self.queue.tasks.get_task(rtxn, task_id)?.ok_or(Error::CorruptedTaskQueue)?;
+            current_batch.processing(Some(&mut task));
+            return Ok(Some((Batch::TaskCancelation { task }, current_batch)));
+        }
+
+        // 2. we get the next task to delete
+        let to_delete = self.queue.tasks.get_kind(rtxn, Kind::TaskDeletion)? & enqueued;
+        if !to_delete.is_empty() {
+            let mut tasks = self.queue.tasks.get_existing_tasks(rtxn, to_delete)?;
+            current_batch.processing(&mut tasks);
+            return Ok(Some((Batch::TaskDeletions(tasks), current_batch)));
+        }
+
+        // 3. we batch the snapshot.
+        let to_snapshot = self.queue.tasks.get_kind(rtxn, Kind::SnapshotCreation)? & enqueued;
+        if !to_snapshot.is_empty() {
+            let mut tasks = self.queue.tasks.get_existing_tasks(rtxn, to_snapshot)?;
+            current_batch.processing(&mut tasks);
+            return Ok(Some((Batch::SnapshotCreation(tasks), current_batch)));
+        }
+
+        // 4. we batch the dumps.
+        let to_dump = self.queue.tasks.get_kind(rtxn, Kind::DumpCreation)? & enqueued;
+        if let Some(to_dump) = to_dump.min() {
+            let mut task =
+                self.queue.tasks.get_task(rtxn, to_dump)?.ok_or(Error::CorruptedTaskQueue)?;
+            current_batch.processing(Some(&mut task));
+            return Ok(Some((Batch::Dump(task), current_batch)));
+        }
+
+        // 5. We make a batch from the unprioritised tasks. Start by taking the next enqueued task.
+        let task_id = if let Some(task_id) = enqueued.min() { task_id } else { return Ok(None) };
+        let mut task =
+            self.queue.tasks.get_task(rtxn, task_id)?.ok_or(Error::CorruptedTaskQueue)?;
+
+        // If the task is not associated with any index, verify that it is an index swap and
+        // create the batch directly. Otherwise, get the index name associated with the task
+        // and use the autobatcher to batch the enqueued tasks associated with it
+
+        let index_name = if let Some(&index_name) = task.indexes().first() {
+            index_name
+        } else {
+            assert!(matches!(&task.kind, KindWithContent::IndexSwap { swaps } if swaps.is_empty()));
+            current_batch.processing(Some(&mut task));
+            return Ok(Some((Batch::IndexSwap { task }, current_batch)));
+        };
+
+        let index_already_exists = self.index_mapper.exists(rtxn, index_name)?;
+        let mut primary_key = None;
+        if index_already_exists {
+            let index = self.index_mapper.index(rtxn, index_name)?;
+            let rtxn = index.read_txn()?;
+            primary_key = index.primary_key(&rtxn)?.map(|pk| pk.to_string());
+        }
+
+        let index_tasks = self.queue.tasks.index_tasks(rtxn, index_name)? & enqueued;
+
+        // If autobatching is disabled we only take one task at a time.
+        // Otherwise, we take only a maximum of tasks to create batches.
+        let tasks_limit = if self.scheduler.autobatching_enabled {
+            self.scheduler.max_number_of_batched_tasks
+        } else {
+            1
+        };
+
+        let mut enqueued = Vec::new();
+        let mut total_size: u64 = 0;
+        for task_id in index_tasks.into_iter().take(tasks_limit) {
+            let task = self
+                .queue
+                .tasks
+                .get_task(rtxn, task_id)
+                .and_then(|task| task.ok_or(Error::CorruptedTaskQueue))?;
+
+            if let Some(uuid) = task.content_uuid() {
+                let content_size = self.queue.file_store.compute_size(uuid)?;
+                total_size = total_size.saturating_add(content_size);
+            }
+
+            if total_size > self.scheduler.batched_tasks_size_limit && !enqueued.is_empty() {
+                break;
+            }
+
+            enqueued.push((task.uid, task.kind));
+        }
+
+        if let Some((batchkind, create_index)) =
+            autobatcher::autobatch(enqueued, index_already_exists, primary_key.as_deref())
+        {
+            return Ok(self
+                .create_next_batch_index(
+                    rtxn,
+                    index_name.to_string(),
+                    batchkind,
+                    &mut current_batch,
+                    create_index,
+                )?
+                .map(|batch| (batch, current_batch)));
+        }
+
+        // If we found no tasks then we were notified for something that got autobatched
+        // somehow and there is nothing to do.
+        Ok(None)
+    }
+}

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

@@ -0,0 +1,453 @@
+mod autobatcher;
+#[cfg(test)]
+mod autobatcher_test;
+mod create_batch;
+mod process_batch;
+mod process_dump_creation;
+mod process_index_operation;
+mod process_snapshot_creation;
+mod process_upgrade;
+#[cfg(test)]
+mod test;
+#[cfg(test)]
+mod test_document_addition;
+#[cfg(test)]
+mod test_embedders;
+#[cfg(test)]
+mod test_failure;
+
+use std::path::PathBuf;
+use std::sync::atomic::{AtomicBool, AtomicU32, Ordering};
+use std::sync::Arc;
+
+use convert_case::{Case, Casing as _};
+use meilisearch_types::error::ResponseError;
+use meilisearch_types::heed::{Env, WithoutTls};
+use meilisearch_types::milli;
+use meilisearch_types::tasks::Status;
+use process_batch::ProcessBatchInfo;
+use rayon::current_num_threads;
+use rayon::iter::{IntoParallelIterator, ParallelIterator};
+use roaring::RoaringBitmap;
+use synchronoise::SignalEvent;
+
+use crate::processing::{AtomicTaskStep, BatchProgress};
+use crate::{Error, IndexScheduler, IndexSchedulerOptions, Result, TickOutcome};
+
+#[derive(Default, Clone, Debug)]
+pub struct MustStopProcessing(Arc<AtomicBool>);
+
+impl MustStopProcessing {
+    pub fn get(&self) -> bool {
+        self.0.load(Ordering::Relaxed)
+    }
+
+    pub fn must_stop(&self) {
+        self.0.store(true, Ordering::Relaxed);
+    }
+
+    pub fn reset(&self) {
+        self.0.store(false, Ordering::Relaxed);
+    }
+}
+
+pub struct Scheduler {
+    /// A boolean that can be set to true to stop the currently processing tasks.
+    pub must_stop_processing: MustStopProcessing,
+
+    /// Get a signal when a batch needs to be processed.
+    pub(crate) wake_up: Arc<SignalEvent>,
+
+    /// Whether auto-batching is enabled or not.
+    pub(crate) autobatching_enabled: bool,
+
+    /// The maximum number of tasks that will be batched together.
+    pub(crate) max_number_of_batched_tasks: usize,
+
+    /// The maximum size, in bytes, of tasks in a batch.
+    pub(crate) batched_tasks_size_limit: u64,
+
+    /// The path used to create the dumps.
+    pub(crate) dumps_path: PathBuf,
+
+    /// The path used to create the snapshots.
+    pub(crate) snapshots_path: PathBuf,
+
+    /// The path to the folder containing the auth LMDB env.
+    pub(crate) auth_env: Env<WithoutTls>,
+
+    /// The path to the version file of Meilisearch.
+    pub(crate) version_file_path: PathBuf,
+
+    /// The maximal number of entries in the search query cache of an embedder.
+    ///
+    /// 0 disables the cache.
+    pub(crate) embedding_cache_cap: usize,
+}
+
+impl Scheduler {
+    pub(crate) fn private_clone(&self) -> Scheduler {
+        Scheduler {
+            must_stop_processing: self.must_stop_processing.clone(),
+            wake_up: self.wake_up.clone(),
+            autobatching_enabled: self.autobatching_enabled,
+            max_number_of_batched_tasks: self.max_number_of_batched_tasks,
+            batched_tasks_size_limit: self.batched_tasks_size_limit,
+            dumps_path: self.dumps_path.clone(),
+            snapshots_path: self.snapshots_path.clone(),
+            auth_env: self.auth_env.clone(),
+            version_file_path: self.version_file_path.clone(),
+            embedding_cache_cap: self.embedding_cache_cap,
+        }
+    }
+
+    pub fn new(options: &IndexSchedulerOptions, auth_env: Env<WithoutTls>) -> Scheduler {
+        Scheduler {
+            must_stop_processing: MustStopProcessing::default(),
+            // we want to start the loop right away in case meilisearch was ctrl+Ced while processing things
+            wake_up: Arc::new(SignalEvent::auto(true)),
+            autobatching_enabled: options.autobatching_enabled,
+            max_number_of_batched_tasks: options.max_number_of_batched_tasks,
+            batched_tasks_size_limit: options.batched_tasks_size_limit,
+            dumps_path: options.dumps_path.clone(),
+            snapshots_path: options.snapshots_path.clone(),
+            auth_env,
+            version_file_path: options.version_file_path.clone(),
+            embedding_cache_cap: options.embedding_cache_cap,
+        }
+    }
+}
+
+impl IndexScheduler {
+    /// Perform one iteration of the run loop.
+    ///
+    /// 1. See if we need to cleanup the task queue
+    /// 2. Find the next batch of tasks to be processed.
+    /// 3. Update the information of these tasks following the start of their processing.
+    /// 4. Update the in-memory list of processed tasks accordingly.
+    /// 5. Process the batch:
+    ///    - perform the actions of each batched task
+    ///    - update the information of each batched task following the end
+    ///      of their processing.
+    /// 6. Reset the in-memory list of processed tasks.
+    ///
+    /// Returns the number of processed tasks.
+    pub(crate) fn tick(&self) -> Result<TickOutcome> {
+        #[cfg(test)]
+        {
+            *self.run_loop_iteration.write().unwrap() += 1;
+            self.breakpoint(crate::test_utils::Breakpoint::Start);
+        }
+
+        if self.cleanup_enabled {
+            let mut wtxn = self.env.write_txn()?;
+            self.queue.cleanup_task_queue(&mut wtxn)?;
+            wtxn.commit()?;
+        }
+
+        let rtxn = self.env.read_txn().map_err(Error::HeedTransaction)?;
+        let (batch, mut processing_batch) =
+            match self.create_next_batch(&rtxn).map_err(|e| Error::CreateBatch(Box::new(e)))? {
+                Some(batch) => batch,
+                None => return Ok(TickOutcome::WaitForSignal),
+            };
+        let index_uid = batch.index_uid().map(ToOwned::to_owned);
+        drop(rtxn);
+
+        // 1. store the starting date with the bitmap of processing tasks.
+        let mut ids = batch.ids();
+        let processed_tasks = ids.len();
+
+        // We reset the must_stop flag to be sure that we don't stop processing tasks
+        self.scheduler.must_stop_processing.reset();
+        let progress = self
+            .processing_tasks
+            .write()
+            .unwrap()
+            // We can clone the processing batch here because we don't want its modification to affect the view of the processing batches
+            .start_processing(processing_batch.clone(), ids.clone());
+
+        #[cfg(test)]
+        self.breakpoint(crate::test_utils::Breakpoint::BatchCreated);
+
+        // 2. Process the tasks
+        let res = {
+            let cloned_index_scheduler = self.private_clone();
+            let processing_batch = &mut processing_batch;
+            let progress = progress.clone();
+            std::thread::scope(|s| {
+                let p = progress.clone();
+                let handle = std::thread::Builder::new()
+                    .name(String::from("batch-operation"))
+                    .spawn_scoped(s, move || {
+                        cloned_index_scheduler.process_batch(batch, processing_batch, p)
+                    })
+                    .unwrap();
+
+                match handle.join() {
+                    Ok(ret) => {
+                        if ret.is_err() {
+                            if let Ok(progress_view) =
+                                serde_json::to_string(&progress.as_progress_view())
+                            {
+                                tracing::warn!("Batch failed while doing: {progress_view}")
+                            }
+                        }
+                        ret
+                    }
+                    Err(panic) => {
+                        if let Ok(progress_view) =
+                            serde_json::to_string(&progress.as_progress_view())
+                        {
+                            tracing::warn!("Batch failed while doing: {progress_view}")
+                        }
+                        let msg = match panic.downcast_ref::<&'static str>() {
+                            Some(s) => *s,
+                            None => match panic.downcast_ref::<String>() {
+                                Some(s) => &s[..],
+                                None => "Box<dyn Any>",
+                            },
+                        };
+                        Err(Error::ProcessBatchPanicked(msg.to_string()))
+                    }
+                }
+            })
+        };
+
+        // Reset the currently updating index to relinquish the index handle
+        self.index_mapper.set_currently_updating_index(None);
+
+        #[cfg(test)]
+        self.maybe_fail(crate::test_utils::FailureLocation::AcquiringWtxn)?;
+
+        progress.update_progress(BatchProgress::WritingTasksToDisk);
+        processing_batch.finished();
+        let mut stop_scheduler_forever = false;
+        let mut wtxn = self.env.write_txn().map_err(Error::HeedTransaction)?;
+        let mut canceled = RoaringBitmap::new();
+        let mut process_batch_info = ProcessBatchInfo::default();
+
+        match res {
+            Ok((tasks, info)) => {
+                #[cfg(test)]
+                self.breakpoint(crate::test_utils::Breakpoint::ProcessBatchSucceeded);
+
+                let (task_progress, task_progress_obj) = AtomicTaskStep::new(tasks.len() as u32);
+                progress.update_progress(task_progress_obj);
+                process_batch_info = info;
+                let mut success = 0;
+                let mut failure = 0;
+                let mut canceled_by = None;
+
+                #[allow(unused_variables)]
+                for (i, mut task) in tasks.into_iter().enumerate() {
+                    task_progress.fetch_add(1, Ordering::Relaxed);
+                    processing_batch.update(&mut task);
+                    if task.status == Status::Canceled {
+                        canceled.insert(task.uid);
+                        canceled_by = task.canceled_by;
+                    }
+
+                    #[cfg(test)]
+                    self.maybe_fail(
+                        crate::test_utils::FailureLocation::UpdatingTaskAfterProcessBatchSuccess {
+                            task_uid: i as u32,
+                        },
+                    )?;
+
+                    match task.error {
+                        Some(_) => failure += 1,
+                        None => success += 1,
+                    }
+
+                    self.queue
+                        .tasks
+                        .update_task(&mut wtxn, &task)
+                        .map_err(|e| Error::UnrecoverableError(Box::new(e)))?;
+                }
+                if let Some(canceled_by) = canceled_by {
+                    self.queue.tasks.canceled_by.put(&mut wtxn, &canceled_by, &canceled)?;
+                }
+                tracing::info!("A batch of tasks was successfully completed with {success} successful tasks and {failure} failed tasks.");
+            }
+            // If we have an abortion error we must stop the tick here and re-schedule tasks.
+            Err(Error::Milli {
+                error: milli::Error::InternalError(milli::InternalError::AbortedIndexation),
+                ..
+            })
+            | Err(Error::AbortedTask) => {
+                #[cfg(test)]
+                self.breakpoint(crate::test_utils::Breakpoint::AbortedIndexation);
+                wtxn.abort();
+
+                tracing::info!("A batch of tasks was aborted.");
+                // We make sure that we don't call `stop_processing` on the `processing_tasks`,
+                // this is because we want to let the next tick call `create_next_batch` and keep
+                // the `started_at` date times and `processings` of the current processing tasks.
+                // This date time is used by the task cancelation to store the right `started_at`
+                // date in the task on disk.
+                return Ok(TickOutcome::TickAgain(0));
+            }
+            // If an index said it was full, we need to:
+            // 1. identify which index is full
+            // 2. close the associated environment
+            // 3. resize it
+            // 4. re-schedule tasks
+            Err(Error::Milli {
+                error: milli::Error::UserError(milli::UserError::MaxDatabaseSizeReached),
+                ..
+            }) if index_uid.is_some() => {
+                // fixme: add index_uid to match to avoid the unwrap
+                let index_uid = index_uid.unwrap();
+                // fixme: handle error more gracefully? not sure when this could happen
+                self.index_mapper.resize_index(&wtxn, &index_uid)?;
+                wtxn.abort();
+
+                tracing::info!("The max database size was reached. Resizing the index.");
+
+                return Ok(TickOutcome::TickAgain(0));
+            }
+            // In case of a failure we must get back and patch all the tasks with the error.
+            Err(err) => {
+                #[cfg(test)]
+                self.breakpoint(crate::test_utils::Breakpoint::ProcessBatchFailed);
+                let (task_progress, task_progress_obj) = AtomicTaskStep::new(ids.len() as u32);
+                progress.update_progress(task_progress_obj);
+
+                if matches!(err, Error::DatabaseUpgrade(_)) {
+                    tracing::error!(
+                        "Upgrade task failed, tasks won't be processed until the following issue is fixed: {err}"
+                    );
+                    stop_scheduler_forever = true;
+                }
+                let error: ResponseError = err.into();
+                for id in ids.iter() {
+                    task_progress.fetch_add(1, Ordering::Relaxed);
+                    let mut task = self
+                        .queue
+                        .tasks
+                        .get_task(&wtxn, id)
+                        .map_err(|e| Error::UnrecoverableError(Box::new(e)))?
+                        .ok_or(Error::CorruptedTaskQueue)?;
+                    task.status = Status::Failed;
+                    task.error = Some(error.clone());
+                    task.details = task.details.map(|d| d.to_failed());
+                    processing_batch.update(&mut task);
+
+                    #[cfg(test)]
+                    self.maybe_fail(
+                        crate::test_utils::FailureLocation::UpdatingTaskAfterProcessBatchFailure,
+                    )?;
+
+                    tracing::error!("Batch failed {}", error);
+
+                    self.queue
+                        .tasks
+                        .update_task(&mut wtxn, &task)
+                        .map_err(|e| Error::UnrecoverableError(Box::new(e)))?;
+                }
+            }
+        }
+
+        // We must re-add the canceled task so they're part of the same batch.
+        ids |= canceled;
+
+        let ProcessBatchInfo { congestion, pre_commit_dabases_sizes, post_commit_dabases_sizes } =
+            process_batch_info;
+
+        processing_batch.stats.progress_trace =
+            progress.accumulated_durations().into_iter().map(|(k, v)| (k, v.into())).collect();
+        processing_batch.stats.write_channel_congestion = congestion.map(|congestion| {
+            let mut congestion_info = serde_json::Map::new();
+            congestion_info.insert("attempts".into(), congestion.attempts.into());
+            congestion_info.insert("blocking_attempts".into(), congestion.blocking_attempts.into());
+            congestion_info.insert("blocking_ratio".into(), congestion.congestion_ratio().into());
+            congestion_info
+        });
+        processing_batch.stats.internal_database_sizes = pre_commit_dabases_sizes
+            .iter()
+            .flat_map(|(dbname, pre_size)| {
+                post_commit_dabases_sizes
+                    .get(dbname)
+                    .map(|post_size| {
+                        use byte_unit::{Byte, UnitType::Binary};
+                        use std::cmp::Ordering::{Equal, Greater, Less};
+
+                        let post = Byte::from_u64(*post_size as u64).get_appropriate_unit(Binary);
+                        let diff_size = post_size.abs_diff(*pre_size) as u64;
+                        let diff = Byte::from_u64(diff_size).get_appropriate_unit(Binary);
+                        let sign = match post_size.cmp(pre_size) {
+                            Equal => return None,
+                            Greater => "+",
+                            Less => "-",
+                        };
+
+                        Some((
+                            dbname.to_case(Case::Camel),
+                            format!("{post:#.2} ({sign}{diff:#.2})").into(),
+                        ))
+                    })
+                    .into_iter()
+                    .flatten()
+            })
+            .collect();
+
+        if let Some(congestion) = congestion {
+            tracing::debug!(
+                "Channel congestion metrics - Attempts: {}, Blocked attempts: {}  ({:.1}% congestion)",
+                congestion.attempts,
+                congestion.blocking_attempts,
+                congestion.congestion_ratio(),
+            );
+        }
+
+        tracing::debug!("call trace: {:?}", progress.accumulated_durations());
+
+        self.queue.write_batch(&mut wtxn, processing_batch, &ids)?;
+
+        #[cfg(test)]
+        self.maybe_fail(crate::test_utils::FailureLocation::CommittingWtxn)?;
+
+        wtxn.commit().map_err(Error::HeedTransaction)?;
+
+        // We should stop processing AFTER everything is processed and written to disk otherwise, a batch (which only lives in RAM) may appear in the processing task
+        // and then become « not found » for some time until the commit everything is written and the final commit is made.
+        self.processing_tasks.write().unwrap().stop_processing();
+
+        // Once the tasks are committed, we should delete all the update files associated ASAP to avoid leaking files in case of a restart
+        tracing::debug!("Deleting the update files");
+
+        //We take one read transaction **per thread**. Then, every thread is going to pull out new IDs from the roaring bitmap with the help of an atomic shared index into the bitmap
+        let idx = AtomicU32::new(0);
+        (0..current_num_threads()).into_par_iter().try_for_each(|_| -> Result<()> {
+            let rtxn = self.read_txn()?;
+            while let Some(id) = ids.select(idx.fetch_add(1, Ordering::Relaxed)) {
+                let task = self
+                    .queue
+                    .tasks
+                    .get_task(&rtxn, id)
+                    .map_err(|e| Error::UnrecoverableError(Box::new(e)))?
+                    .ok_or(Error::CorruptedTaskQueue)?;
+                if let Err(e) = self.queue.delete_persisted_task_data(&task) {
+                    tracing::error!(
+                        "Failure to delete the content files associated with task {}. Error: {e}",
+                        task.uid
+                    );
+                }
+            }
+            Ok(())
+        })?;
+
+        // We shouldn't crash the tick function if we can't send data to the webhook.
+        let _ = self.notify_webhook(&ids);
+
+        #[cfg(test)]
+        self.breakpoint(crate::test_utils::Breakpoint::AfterProcessing);
+
+        if stop_scheduler_forever {
+            Ok(TickOutcome::StopProcessingForever)
+        } else {
+            Ok(TickOutcome::TickAgain(processed_tasks))
+        }
+    }
+}

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

@@ -0,0 +1,687 @@
+use std::collections::{BTreeSet, HashMap, HashSet};
+use std::panic::{catch_unwind, AssertUnwindSafe};
+use std::sync::atomic::Ordering;
+
+use meilisearch_types::batches::{BatchEnqueuedAt, BatchId};
+use meilisearch_types::heed::{RoTxn, RwTxn};
+use meilisearch_types::milli::progress::{Progress, VariableNameStep};
+use meilisearch_types::milli::{self, ChannelCongestion};
+use meilisearch_types::tasks::{Details, IndexSwap, KindWithContent, Status, Task};
+use milli::update::Settings as MilliSettings;
+use roaring::RoaringBitmap;
+
+use super::create_batch::Batch;
+use crate::processing::{
+    AtomicBatchStep, AtomicTaskStep, CreateIndexProgress, DeleteIndexProgress, FinalizingIndexStep,
+    InnerSwappingTwoIndexes, SwappingTheIndexes, TaskCancelationProgress, TaskDeletionProgress,
+    UpdateIndexProgress,
+};
+use crate::utils::{
+    self, remove_n_tasks_datetime_earlier_than, remove_task_datetime, swap_index_uid_in_task,
+    ProcessingBatch,
+};
+use crate::{Error, IndexScheduler, Result, TaskId};
+
+#[derive(Debug, Default)]
+pub struct ProcessBatchInfo {
+    /// The write channel congestion. None when unavailable: settings update.
+    pub congestion: Option<ChannelCongestion>,
+    /// The sizes of the different databases before starting the indexation.
+    pub pre_commit_dabases_sizes: indexmap::IndexMap<&'static str, usize>,
+    /// The sizes of the different databases after commiting the indexation.
+    pub post_commit_dabases_sizes: indexmap::IndexMap<&'static str, usize>,
+}
+
+impl IndexScheduler {
+    /// Apply the operation associated with the given batch.
+    ///
+    /// ## Return
+    /// The list of tasks that were processed. The metadata of each task in the returned
+    /// list is updated accordingly, with the exception of the its date fields
+    /// [`finished_at`](meilisearch_types::tasks::Task::finished_at) and [`started_at`](meilisearch_types::tasks::Task::started_at).
+    #[tracing::instrument(level = "trace", skip(self, batch, progress), target = "indexing::scheduler", fields(batch=batch.to_string()))]
+    pub(crate) fn process_batch(
+        &self,
+        batch: Batch,
+        current_batch: &mut ProcessingBatch,
+        progress: Progress,
+    ) -> Result<(Vec<Task>, ProcessBatchInfo)> {
+        #[cfg(test)]
+        {
+            self.maybe_fail(crate::test_utils::FailureLocation::InsideProcessBatch)?;
+            self.maybe_fail(crate::test_utils::FailureLocation::PanicInsideProcessBatch)?;
+            self.breakpoint(crate::test_utils::Breakpoint::InsideProcessBatch);
+        }
+
+        match batch {
+            Batch::TaskCancelation { mut task } => {
+                // 1. Retrieve the tasks that matched the query at enqueue-time.
+                let matched_tasks =
+                    if let KindWithContent::TaskCancelation { tasks, query: _ } = &task.kind {
+                        tasks
+                    } else {
+                        unreachable!()
+                    };
+
+                let rtxn = self.env.read_txn()?;
+                let mut canceled_tasks = self.cancel_matched_tasks(
+                    &rtxn,
+                    task.uid,
+                    current_batch,
+                    matched_tasks,
+                    &progress,
+                )?;
+
+                task.status = Status::Succeeded;
+                match &mut task.details {
+                    Some(Details::TaskCancelation {
+                        matched_tasks: _,
+                        canceled_tasks: canceled_tasks_details,
+                        original_filter: _,
+                    }) => {
+                        *canceled_tasks_details = Some(canceled_tasks.len() as u64);
+                    }
+                    _ => unreachable!(),
+                }
+
+                canceled_tasks.push(task);
+
+                Ok((canceled_tasks, ProcessBatchInfo::default()))
+            }
+            Batch::TaskDeletions(mut tasks) => {
+                // 1. Retrieve the tasks that matched the query at enqueue-time.
+                let mut matched_tasks = RoaringBitmap::new();
+
+                for task in tasks.iter() {
+                    if let KindWithContent::TaskDeletion { tasks, query: _ } = &task.kind {
+                        matched_tasks |= tasks;
+                    } else {
+                        unreachable!()
+                    }
+                }
+
+                let mut wtxn = self.env.write_txn()?;
+                let mut deleted_tasks =
+                    self.delete_matched_tasks(&mut wtxn, &matched_tasks, &progress)?;
+                wtxn.commit()?;
+
+                for task in tasks.iter_mut() {
+                    task.status = Status::Succeeded;
+                    let KindWithContent::TaskDeletion { tasks, query: _ } = &task.kind else {
+                        unreachable!()
+                    };
+
+                    let deleted_tasks_count = deleted_tasks.intersection_len(tasks);
+                    deleted_tasks -= tasks;
+
+                    match &mut task.details {
+                        Some(Details::TaskDeletion {
+                            matched_tasks: _,
+                            deleted_tasks,
+                            original_filter: _,
+                        }) => {
+                            *deleted_tasks = Some(deleted_tasks_count);
+                        }
+                        _ => unreachable!(),
+                    }
+                }
+                Ok((tasks, ProcessBatchInfo::default()))
+            }
+            Batch::SnapshotCreation(tasks) => self
+                .process_snapshot(progress, tasks)
+                .map(|tasks| (tasks, ProcessBatchInfo::default())),
+            Batch::Dump(task) => self
+                .process_dump_creation(progress, task)
+                .map(|tasks| (tasks, ProcessBatchInfo::default())),
+            Batch::IndexOperation { op, must_create_index } => {
+                let index_uid = op.index_uid().to_string();
+                let index = if must_create_index {
+                    // create the index if it doesn't already exist
+                    let wtxn = self.env.write_txn()?;
+                    self.index_mapper.create_index(wtxn, &index_uid, None)?
+                } else {
+                    let rtxn = self.env.read_txn()?;
+                    self.index_mapper.index(&rtxn, &index_uid)?
+                };
+
+                // the index operation can take a long time, so save this handle to make it available to the search for the duration of the tick
+                self.index_mapper
+                    .set_currently_updating_index(Some((index_uid.clone(), index.clone())));
+
+                let mut index_wtxn = index.write_txn()?;
+                let pre_commit_dabases_sizes = index.database_sizes(&index_wtxn)?;
+                let (tasks, congestion) =
+                    self.apply_index_operation(&mut index_wtxn, &index, op, &progress)?;
+
+                {
+                    progress.update_progress(FinalizingIndexStep::Committing);
+                    let span = tracing::trace_span!(target: "indexing::scheduler", "commit");
+                    let _entered = span.enter();
+
+                    index_wtxn.commit()?;
+                }
+
+                // if the update processed successfully, we're going to store the new
+                // stats of the index. Since the tasks have already been processed and
+                // this is a non-critical operation. If it fails, we should not fail
+                // the entire batch.
+                let mut post_commit_dabases_sizes = None;
+                let res = || -> Result<()> {
+                    progress.update_progress(FinalizingIndexStep::ComputingStats);
+                    let index_rtxn = index.read_txn()?;
+                    let stats = crate::index_mapper::IndexStats::new(&index, &index_rtxn)
+                        .map_err(|e| Error::from_milli(e, Some(index_uid.to_string())))?;
+                    let mut wtxn = self.env.write_txn()?;
+                    self.index_mapper.store_stats_of(&mut wtxn, &index_uid, &stats)?;
+                    post_commit_dabases_sizes = Some(index.database_sizes(&index_rtxn)?);
+                    wtxn.commit()?;
+                    Ok(())
+                }();
+
+                match res {
+                    Ok(_) => (),
+                    Err(e) => tracing::error!(
+                        error = &e as &dyn std::error::Error,
+                        "Could not write the stats of the index"
+                    ),
+                }
+
+                let info = ProcessBatchInfo {
+                    congestion,
+                    // In case we fail to the get post-commit sizes we decide
+                    // that nothing changed and use the pre-commit sizes.
+                    post_commit_dabases_sizes: post_commit_dabases_sizes
+                        .unwrap_or_else(|| pre_commit_dabases_sizes.clone()),
+                    pre_commit_dabases_sizes,
+                };
+
+                Ok((tasks, info))
+            }
+            Batch::IndexCreation { index_uid, primary_key, task } => {
+                progress.update_progress(CreateIndexProgress::CreatingTheIndex);
+
+                let wtxn = self.env.write_txn()?;
+                if self.index_mapper.exists(&wtxn, &index_uid)? {
+                    return Err(Error::IndexAlreadyExists(index_uid));
+                }
+                self.index_mapper.create_index(wtxn, &index_uid, None)?;
+
+                self.process_batch(
+                    Batch::IndexUpdate { index_uid, primary_key, task },
+                    current_batch,
+                    progress,
+                )
+            }
+            Batch::IndexUpdate { index_uid, primary_key, mut task } => {
+                progress.update_progress(UpdateIndexProgress::UpdatingTheIndex);
+                let rtxn = self.env.read_txn()?;
+                let index = self.index_mapper.index(&rtxn, &index_uid)?;
+
+                if let Some(primary_key) = primary_key.clone() {
+                    let mut index_wtxn = index.write_txn()?;
+                    let mut builder = MilliSettings::new(
+                        &mut index_wtxn,
+                        &index,
+                        self.index_mapper.indexer_config(),
+                    );
+                    builder.set_primary_key(primary_key);
+                    let must_stop_processing = self.scheduler.must_stop_processing.clone();
+                    builder
+                        .execute(
+                            |indexing_step| tracing::debug!(update = ?indexing_step),
+                            || must_stop_processing.get(),
+                        )
+                        .map_err(|e| Error::from_milli(e, Some(index_uid.to_string())))?;
+                    index_wtxn.commit()?;
+                }
+
+                // drop rtxn before starting a new wtxn on the same db
+                rtxn.commit()?;
+
+                task.status = Status::Succeeded;
+                task.details = Some(Details::IndexInfo { primary_key });
+
+                // if the update processed successfully, we're going to store the new
+                // stats of the index. Since the tasks have already been processed and
+                // this is a non-critical operation. If it fails, we should not fail
+                // the entire batch.
+                let res = || -> Result<()> {
+                    let mut wtxn = self.env.write_txn()?;
+                    let index_rtxn = index.read_txn()?;
+                    let stats = crate::index_mapper::IndexStats::new(&index, &index_rtxn)
+                        .map_err(|e| Error::from_milli(e, Some(index_uid.clone())))?;
+                    self.index_mapper.store_stats_of(&mut wtxn, &index_uid, &stats)?;
+                    wtxn.commit()?;
+                    Ok(())
+                }();
+
+                match res {
+                    Ok(_) => (),
+                    Err(e) => tracing::error!(
+                        error = &e as &dyn std::error::Error,
+                        "Could not write the stats of the index"
+                    ),
+                }
+
+                Ok((vec![task], ProcessBatchInfo::default()))
+            }
+            Batch::IndexDeletion { index_uid, index_has_been_created, mut tasks } => {
+                progress.update_progress(DeleteIndexProgress::DeletingTheIndex);
+                let wtxn = self.env.write_txn()?;
+
+                // it's possible that the index doesn't exist
+                let number_of_documents = || -> Result<u64> {
+                    let index = self.index_mapper.index(&wtxn, &index_uid)?;
+                    let index_rtxn = index.read_txn()?;
+                    index
+                        .number_of_documents(&index_rtxn)
+                        .map_err(|e| Error::from_milli(e, Some(index_uid.to_string())))
+                }()
+                .unwrap_or_default();
+
+                // The write transaction is directly owned and committed inside.
+                match self.index_mapper.delete_index(wtxn, &index_uid) {
+                    Ok(()) => (),
+                    Err(Error::IndexNotFound(_)) if index_has_been_created => (),
+                    Err(e) => return Err(e),
+                }
+
+                // We set all the tasks details to the default value.
+                for task in &mut tasks {
+                    task.status = Status::Succeeded;
+                    task.details = match &task.kind {
+                        KindWithContent::IndexDeletion { .. } => {
+                            Some(Details::ClearAll { deleted_documents: Some(number_of_documents) })
+                        }
+                        otherwise => otherwise.default_finished_details(),
+                    };
+                }
+
+                // Here we could also show that all the internal database sizes goes to 0
+                // but it would mean opening the index and that's costly.
+                Ok((tasks, ProcessBatchInfo::default()))
+            }
+            Batch::IndexSwap { mut task } => {
+                progress.update_progress(SwappingTheIndexes::EnsuringCorrectnessOfTheSwap);
+
+                let mut wtxn = self.env.write_txn()?;
+                let swaps = if let KindWithContent::IndexSwap { swaps } = &task.kind {
+                    swaps
+                } else {
+                    unreachable!()
+                };
+                let mut not_found_indexes = BTreeSet::new();
+                for IndexSwap { indexes: (lhs, rhs) } in swaps {
+                    for index in [lhs, rhs] {
+                        let index_exists = self.index_mapper.index_exists(&wtxn, index)?;
+                        if !index_exists {
+                            not_found_indexes.insert(index);
+                        }
+                    }
+                }
+                if !not_found_indexes.is_empty() {
+                    if not_found_indexes.len() == 1 {
+                        return Err(Error::SwapIndexNotFound(
+                            not_found_indexes.into_iter().next().unwrap().clone(),
+                        ));
+                    } else {
+                        return Err(Error::SwapIndexesNotFound(
+                            not_found_indexes.into_iter().cloned().collect(),
+                        ));
+                    }
+                }
+                progress.update_progress(SwappingTheIndexes::SwappingTheIndexes);
+                for (step, swap) in swaps.iter().enumerate() {
+                    progress.update_progress(VariableNameStep::<SwappingTheIndexes>::new(
+                        format!("swapping index {} and {}", swap.indexes.0, swap.indexes.1),
+                        step as u32,
+                        swaps.len() as u32,
+                    ));
+                    self.apply_index_swap(
+                        &mut wtxn,
+                        &progress,
+                        task.uid,
+                        &swap.indexes.0,
+                        &swap.indexes.1,
+                    )?;
+                }
+                wtxn.commit()?;
+                task.status = Status::Succeeded;
+                Ok((vec![task], ProcessBatchInfo::default()))
+            }
+            Batch::UpgradeDatabase { mut tasks } => {
+                let KindWithContent::UpgradeDatabase { from } = tasks.last().unwrap().kind else {
+                    unreachable!();
+                };
+                let ret = catch_unwind(AssertUnwindSafe(|| self.process_upgrade(from, progress)));
+                match ret {
+                    Ok(Ok(())) => (),
+                    Ok(Err(e)) => return Err(Error::DatabaseUpgrade(Box::new(e))),
+                    Err(e) => {
+                        let msg = match e.downcast_ref::<&'static str>() {
+                            Some(s) => *s,
+                            None => match e.downcast_ref::<String>() {
+                                Some(s) => &s[..],
+                                None => "Box<dyn Any>",
+                            },
+                        };
+                        return Err(Error::DatabaseUpgrade(Box::new(Error::ProcessBatchPanicked(
+                            msg.to_string(),
+                        ))));
+                    }
+                }
+
+                for task in tasks.iter_mut() {
+                    task.status = Status::Succeeded;
+                    // Since this task can be retried we must reset its error status
+                    task.error = None;
+                }
+
+                Ok((tasks, ProcessBatchInfo::default()))
+            }
+        }
+    }
+
+    /// Swap the index `lhs` with the index `rhs`.
+    fn apply_index_swap(
+        &self,
+        wtxn: &mut RwTxn,
+        progress: &Progress,
+        task_id: u32,
+        lhs: &str,
+        rhs: &str,
+    ) -> Result<()> {
+        progress.update_progress(InnerSwappingTwoIndexes::RetrieveTheTasks);
+        // 1. Verify that both lhs and rhs are existing indexes
+        let index_lhs_exists = self.index_mapper.index_exists(wtxn, lhs)?;
+        if !index_lhs_exists {
+            return Err(Error::IndexNotFound(lhs.to_owned()));
+        }
+        let index_rhs_exists = self.index_mapper.index_exists(wtxn, rhs)?;
+        if !index_rhs_exists {
+            return Err(Error::IndexNotFound(rhs.to_owned()));
+        }
+
+        // 2. Get the task set for index = name that appeared before the index swap task
+        let mut index_lhs_task_ids = self.queue.tasks.index_tasks(wtxn, lhs)?;
+        index_lhs_task_ids.remove_range(task_id..);
+        let mut index_rhs_task_ids = self.queue.tasks.index_tasks(wtxn, rhs)?;
+        index_rhs_task_ids.remove_range(task_id..);
+
+        // 3. before_name -> new_name in the task's KindWithContent
+        progress.update_progress(InnerSwappingTwoIndexes::UpdateTheTasks);
+        let tasks_to_update = &index_lhs_task_ids | &index_rhs_task_ids;
+        let (atomic, task_progress) = AtomicTaskStep::new(tasks_to_update.len() as u32);
+        progress.update_progress(task_progress);
+
+        for task_id in tasks_to_update {
+            let mut task =
+                self.queue.tasks.get_task(wtxn, task_id)?.ok_or(Error::CorruptedTaskQueue)?;
+            swap_index_uid_in_task(&mut task, (lhs, rhs));
+            self.queue.tasks.all_tasks.put(wtxn, &task_id, &task)?;
+            atomic.fetch_add(1, Ordering::Relaxed);
+        }
+
+        // 4. remove the task from indexuid = before_name
+        // 5. add the task to indexuid = after_name
+        progress.update_progress(InnerSwappingTwoIndexes::UpdateTheIndexesMetadata);
+        self.queue.tasks.update_index(wtxn, lhs, |lhs_tasks| {
+            *lhs_tasks -= &index_lhs_task_ids;
+            *lhs_tasks |= &index_rhs_task_ids;
+        })?;
+        self.queue.tasks.update_index(wtxn, rhs, |rhs_tasks| {
+            *rhs_tasks -= &index_rhs_task_ids;
+            *rhs_tasks |= &index_lhs_task_ids;
+        })?;
+
+        // 6. Swap in the index mapper
+        self.index_mapper.swap(wtxn, lhs, rhs)?;
+
+        Ok(())
+    }
+
+    /// Delete each given task from all the databases (if it is deleteable).
+    ///
+    /// Return the number of tasks that were actually deleted.
+    fn delete_matched_tasks(
+        &self,
+        wtxn: &mut RwTxn,
+        matched_tasks: &RoaringBitmap,
+        progress: &Progress,
+    ) -> Result<RoaringBitmap> {
+        progress.update_progress(TaskDeletionProgress::DeletingTasksDateTime);
+
+        // 1. Remove from this list the tasks that we are not allowed to delete
+        let enqueued_tasks = self.queue.tasks.get_status(wtxn, Status::Enqueued)?;
+        let processing_tasks = &self.processing_tasks.read().unwrap().processing.clone();
+
+        let all_task_ids = self.queue.tasks.all_task_ids(wtxn)?;
+        let mut to_delete_tasks = all_task_ids & matched_tasks;
+        to_delete_tasks -= &**processing_tasks;
+        to_delete_tasks -= &enqueued_tasks;
+
+        // 2. We now have a list of tasks to delete, delete them
+        let mut affected_indexes = HashSet::new();
+        let mut affected_statuses = HashSet::new();
+        let mut affected_kinds = HashSet::new();
+        let mut affected_canceled_by = RoaringBitmap::new();
+        // The tasks that have been removed *per batches*.
+        let mut affected_batches: HashMap<BatchId, RoaringBitmap> = HashMap::new();
+
+        let (atomic_progress, task_progress) = AtomicTaskStep::new(to_delete_tasks.len() as u32);
+        progress.update_progress(task_progress);
+        for task_id in to_delete_tasks.iter() {
+            let task =
+                self.queue.tasks.get_task(wtxn, task_id)?.ok_or(Error::CorruptedTaskQueue)?;
+
+            affected_indexes.extend(task.indexes().into_iter().map(|x| x.to_owned()));
+            affected_statuses.insert(task.status);
+            affected_kinds.insert(task.kind.as_kind());
+            // Note: don't delete the persisted task data since
+            // we can only delete succeeded, failed, and canceled tasks.
+            // In each of those cases, the persisted data is supposed to
+            // have been deleted already.
+            utils::remove_task_datetime(
+                wtxn,
+                self.queue.tasks.enqueued_at,
+                task.enqueued_at,
+                task.uid,
+            )?;
+            if let Some(started_at) = task.started_at {
+                utils::remove_task_datetime(
+                    wtxn,
+                    self.queue.tasks.started_at,
+                    started_at,
+                    task.uid,
+                )?;
+            }
+            if let Some(finished_at) = task.finished_at {
+                utils::remove_task_datetime(
+                    wtxn,
+                    self.queue.tasks.finished_at,
+                    finished_at,
+                    task.uid,
+                )?;
+            }
+            if let Some(canceled_by) = task.canceled_by {
+                affected_canceled_by.insert(canceled_by);
+            }
+            if let Some(batch_uid) = task.batch_uid {
+                affected_batches.entry(batch_uid).or_default().insert(task_id);
+            }
+            atomic_progress.fetch_add(1, Ordering::Relaxed);
+        }
+
+        progress.update_progress(TaskDeletionProgress::DeletingTasksMetadata);
+        let (atomic_progress, task_progress) = AtomicTaskStep::new(
+            (affected_indexes.len() + affected_statuses.len() + affected_kinds.len()) as u32,
+        );
+        progress.update_progress(task_progress);
+        for index in affected_indexes.iter() {
+            self.queue.tasks.update_index(wtxn, index, |bitmap| *bitmap -= &to_delete_tasks)?;
+            atomic_progress.fetch_add(1, Ordering::Relaxed);
+        }
+
+        for status in affected_statuses.iter() {
+            self.queue.tasks.update_status(wtxn, *status, |bitmap| *bitmap -= &to_delete_tasks)?;
+            atomic_progress.fetch_add(1, Ordering::Relaxed);
+        }
+
+        for kind in affected_kinds.iter() {
+            self.queue.tasks.update_kind(wtxn, *kind, |bitmap| *bitmap -= &to_delete_tasks)?;
+            atomic_progress.fetch_add(1, Ordering::Relaxed);
+        }
+
+        progress.update_progress(TaskDeletionProgress::DeletingTasks);
+        let (atomic_progress, task_progress) = AtomicTaskStep::new(to_delete_tasks.len() as u32);
+        progress.update_progress(task_progress);
+        for task in to_delete_tasks.iter() {
+            self.queue.tasks.all_tasks.delete(wtxn, &task)?;
+            atomic_progress.fetch_add(1, Ordering::Relaxed);
+        }
+        for canceled_by in affected_canceled_by {
+            if let Some(mut tasks) = self.queue.tasks.canceled_by.get(wtxn, &canceled_by)? {
+                tasks -= &to_delete_tasks;
+                if tasks.is_empty() {
+                    self.queue.tasks.canceled_by.delete(wtxn, &canceled_by)?;
+                } else {
+                    self.queue.tasks.canceled_by.put(wtxn, &canceled_by, &tasks)?;
+                }
+            }
+        }
+        progress.update_progress(TaskDeletionProgress::DeletingBatches);
+        let (atomic_progress, batch_progress) = AtomicBatchStep::new(affected_batches.len() as u32);
+        progress.update_progress(batch_progress);
+        for (batch_id, to_delete_tasks) in affected_batches {
+            if let Some(mut tasks) = self.queue.batch_to_tasks_mapping.get(wtxn, &batch_id)? {
+                tasks -= &to_delete_tasks;
+                // We must remove the batch entirely
+                if tasks.is_empty() {
+                    if let Some(batch) = self.queue.batches.get_batch(wtxn, batch_id)? {
+                        if let Some(BatchEnqueuedAt { earliest, oldest }) = batch.enqueued_at {
+                            remove_task_datetime(
+                                wtxn,
+                                self.queue.batches.enqueued_at,
+                                earliest,
+                                batch_id,
+                            )?;
+                            remove_task_datetime(
+                                wtxn,
+                                self.queue.batches.enqueued_at,
+                                oldest,
+                                batch_id,
+                            )?;
+                        } else {
+                            // If we don't have the enqueued at in the batch it means the database comes from the v1.12
+                            // and we still need to find the date by scrolling the database
+                            remove_n_tasks_datetime_earlier_than(
+                                wtxn,
+                                self.queue.batches.enqueued_at,
+                                batch.started_at,
+                                batch.stats.total_nb_tasks.clamp(1, 2) as usize,
+                                batch_id,
+                            )?;
+                        }
+                        remove_task_datetime(
+                            wtxn,
+                            self.queue.batches.started_at,
+                            batch.started_at,
+                            batch_id,
+                        )?;
+                        if let Some(finished_at) = batch.finished_at {
+                            remove_task_datetime(
+                                wtxn,
+                                self.queue.batches.finished_at,
+                                finished_at,
+                                batch_id,
+                            )?;
+                        }
+
+                        self.queue.batches.all_batches.delete(wtxn, &batch_id)?;
+                        self.queue.batch_to_tasks_mapping.delete(wtxn, &batch_id)?;
+                    }
+                }
+
+                // Anyway, we must remove the batch from all its reverse indexes.
+                // The only way to do that is to check
+
+                for index in affected_indexes.iter() {
+                    let index_tasks = self.queue.tasks.index_tasks(wtxn, index)?;
+                    let remaining_index_tasks = index_tasks & &tasks;
+                    if remaining_index_tasks.is_empty() {
+                        self.queue.batches.update_index(wtxn, index, |bitmap| {
+                            bitmap.remove(batch_id);
+                        })?;
+                    }
+                }
+
+                for status in affected_statuses.iter() {
+                    let status_tasks = self.queue.tasks.get_status(wtxn, *status)?;
+                    let remaining_status_tasks = status_tasks & &tasks;
+                    if remaining_status_tasks.is_empty() {
+                        self.queue.batches.update_status(wtxn, *status, |bitmap| {
+                            bitmap.remove(batch_id);
+                        })?;
+                    }
+                }
+
+                for kind in affected_kinds.iter() {
+                    let kind_tasks = self.queue.tasks.get_kind(wtxn, *kind)?;
+                    let remaining_kind_tasks = kind_tasks & &tasks;
+                    if remaining_kind_tasks.is_empty() {
+                        self.queue.batches.update_kind(wtxn, *kind, |bitmap| {
+                            bitmap.remove(batch_id);
+                        })?;
+                    }
+                }
+            }
+            atomic_progress.fetch_add(1, Ordering::Relaxed);
+        }
+
+        Ok(to_delete_tasks)
+    }
+
+    /// Cancel each given task from all the databases (if it is cancelable).
+    ///
+    /// Returns the list of tasks that matched the filter and must be written in the database.
+    fn cancel_matched_tasks(
+        &self,
+        rtxn: &RoTxn,
+        cancel_task_id: TaskId,
+        current_batch: &mut ProcessingBatch,
+        matched_tasks: &RoaringBitmap,
+        progress: &Progress,
+    ) -> Result<Vec<Task>> {
+        progress.update_progress(TaskCancelationProgress::RetrievingTasks);
+
+        // 1. Remove from this list the tasks that we are not allowed to cancel
+        //    Notice that only the _enqueued_ ones are cancelable and we should
+        //    have already aborted the indexation of the _processing_ ones
+        let cancelable_tasks = self.queue.tasks.get_status(rtxn, Status::Enqueued)?;
+        let tasks_to_cancel = cancelable_tasks & matched_tasks;
+
+        let (task_progress, progress_obj) = AtomicTaskStep::new(tasks_to_cancel.len() as u32);
+        progress.update_progress(progress_obj);
+
+        // 2. We now have a list of tasks to cancel, cancel them
+        let mut tasks = self.queue.tasks.get_existing_tasks(
+            rtxn,
+            tasks_to_cancel.iter().inspect(|_| {
+                task_progress.fetch_add(1, Ordering::Relaxed);
+            }),
+        )?;
+
+        progress.update_progress(TaskCancelationProgress::UpdatingTasks);
+        let (task_progress, progress_obj) = AtomicTaskStep::new(tasks_to_cancel.len() as u32);
+        progress.update_progress(progress_obj);
+        for task in tasks.iter_mut() {
+            task.status = Status::Canceled;
+            task.canceled_by = Some(cancel_task_id);
+            task.details = task.details.as_ref().map(|d| d.to_failed());
+            current_batch.processing(Some(task));
+            task_progress.fetch_add(1, Ordering::Relaxed);
+        }
+
+        Ok(tasks)
+    }
+}

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

@@ -0,0 +1,278 @@
+use std::collections::BTreeMap;
+use std::fs::File;
+use std::io::BufWriter;
+use std::sync::atomic::Ordering;
+
+use dump::IndexMetadata;
+use meilisearch_types::milli::constants::RESERVED_VECTORS_FIELD_NAME;
+use meilisearch_types::milli::progress::{Progress, VariableNameStep};
+use meilisearch_types::milli::vector::parsed_vectors::{ExplicitVectors, VectorOrArrayOfVectors};
+use meilisearch_types::milli::{self};
+use meilisearch_types::tasks::{Details, KindWithContent, Status, Task};
+use time::macros::format_description;
+use time::OffsetDateTime;
+
+use crate::processing::{
+    AtomicBatchStep, AtomicDocumentStep, AtomicTaskStep, DumpCreationProgress,
+};
+use crate::{Error, IndexScheduler, Result};
+
+impl IndexScheduler {
+    pub(super) fn process_dump_creation(
+        &self,
+        progress: Progress,
+        mut task: Task,
+    ) -> Result<Vec<Task>> {
+        progress.update_progress(DumpCreationProgress::StartTheDumpCreation);
+        let started_at = OffsetDateTime::now_utc();
+        let (keys, instance_uid) =
+            if let KindWithContent::DumpCreation { keys, instance_uid } = &task.kind {
+                (keys, instance_uid)
+            } else {
+                unreachable!();
+            };
+        let dump = dump::DumpWriter::new(*instance_uid)?;
+
+        // 1. dump the keys
+        progress.update_progress(DumpCreationProgress::DumpTheApiKeys);
+        let mut dump_keys = dump.create_keys()?;
+        for key in keys {
+            dump_keys.push_key(key)?;
+        }
+        dump_keys.flush()?;
+
+        let rtxn = self.env.read_txn()?;
+
+        // 2. dump the tasks
+        progress.update_progress(DumpCreationProgress::DumpTheTasks);
+        let mut dump_tasks = dump.create_tasks_queue()?;
+
+        let (atomic, update_task_progress) =
+            AtomicTaskStep::new(self.queue.tasks.all_tasks.len(&rtxn)? as u32);
+        progress.update_progress(update_task_progress);
+
+        for ret in self.queue.tasks.all_tasks.iter(&rtxn)? {
+            if self.scheduler.must_stop_processing.get() {
+                return Err(Error::AbortedTask);
+            }
+
+            let (_, mut t) = ret?;
+            let status = t.status;
+            let content_file = t.content_uuid();
+
+            // In the case we're dumping ourselves we want to be marked as finished
+            // to not loop over ourselves indefinitely.
+            if t.uid == task.uid {
+                let finished_at = OffsetDateTime::now_utc();
+
+                // We're going to fake the date because we don't know if everything is going to go well.
+                // But we need to dump the task as finished and successful.
+                // If something fail everything will be set appropriately in the end.
+                t.status = Status::Succeeded;
+                t.started_at = Some(started_at);
+                t.finished_at = Some(finished_at);
+            }
+
+            // Patch the task to remove the batch uid, because as of v1.12.5 batches are not persisted.
+            // This prevent from referencing *future* batches not actually associated with the task.
+            //
+            // See <https://github.com/meilisearch/meilisearch/issues/5247> for details.
+            t.batch_uid = None;
+
+            let mut dump_content_file = dump_tasks.push_task(&t.into())?;
+
+            // 2.1. Dump the `content_file` associated with the task if there is one and the task is not finished yet.
+            if let Some(content_file) = content_file {
+                if self.scheduler.must_stop_processing.get() {
+                    return Err(Error::AbortedTask);
+                }
+                if status == Status::Enqueued {
+                    let content_file = self.queue.file_store.get_update(content_file)?;
+
+                    for document in
+                        serde_json::de::Deserializer::from_reader(content_file).into_iter()
+                    {
+                        let document = document.map_err(|e| {
+                            Error::from_milli(milli::InternalError::SerdeJson(e).into(), None)
+                        })?;
+                        dump_content_file.push_document(&document)?;
+                    }
+
+                    dump_content_file.flush()?;
+                }
+            }
+            atomic.fetch_add(1, Ordering::Relaxed);
+        }
+        dump_tasks.flush()?;
+
+        // 3. dump the batches
+        progress.update_progress(DumpCreationProgress::DumpTheBatches);
+        let mut dump_batches = dump.create_batches_queue()?;
+
+        let (atomic_batch_progress, update_batch_progress) =
+            AtomicBatchStep::new(self.queue.batches.all_batches.len(&rtxn)? as u32);
+        progress.update_progress(update_batch_progress);
+
+        for ret in self.queue.batches.all_batches.iter(&rtxn)? {
+            if self.scheduler.must_stop_processing.get() {
+                return Err(Error::AbortedTask);
+            }
+
+            let (_, mut b) = ret?;
+            // In the case we're dumping ourselves we want to be marked as finished
+            // to not loop over ourselves indefinitely.
+            if b.uid == task.uid {
+                let finished_at = OffsetDateTime::now_utc();
+
+                // We're going to fake the date because we don't know if everything is going to go well.
+                // But we need to dump the task as finished and successful.
+                // If something fail everything will be set appropriately in the end.
+                let mut statuses = BTreeMap::new();
+                statuses.insert(Status::Succeeded, b.stats.total_nb_tasks);
+                b.stats.status = statuses;
+                b.finished_at = Some(finished_at);
+            }
+
+            dump_batches.push_batch(&b)?;
+            atomic_batch_progress.fetch_add(1, Ordering::Relaxed);
+        }
+        dump_batches.flush()?;
+
+        // 4. Dump the indexes
+        progress.update_progress(DumpCreationProgress::DumpTheIndexes);
+        let nb_indexes = self.index_mapper.index_mapping.len(&rtxn)? as u32;
+        let mut count = 0;
+        let () = self.index_mapper.try_for_each_index(&rtxn, |uid, index| -> Result<()> {
+            progress.update_progress(VariableNameStep::<DumpCreationProgress>::new(
+                uid.to_string(),
+                count,
+                nb_indexes,
+            ));
+            count += 1;
+
+            let rtxn = index.read_txn()?;
+            let metadata = IndexMetadata {
+                uid: uid.to_owned(),
+                primary_key: index.primary_key(&rtxn)?.map(String::from),
+                created_at: index
+                    .created_at(&rtxn)
+                    .map_err(|e| Error::from_milli(e, Some(uid.to_string())))?,
+                updated_at: index
+                    .updated_at(&rtxn)
+                    .map_err(|e| Error::from_milli(e, Some(uid.to_string())))?,
+            };
+            let mut index_dumper = dump.create_index(uid, &metadata)?;
+
+            let fields_ids_map = index.fields_ids_map(&rtxn)?;
+            let all_fields: Vec<_> = fields_ids_map.iter().map(|(id, _)| id).collect();
+            let embedding_configs = index
+                .embedding_configs(&rtxn)
+                .map_err(|e| Error::from_milli(e, Some(uid.to_string())))?;
+
+            let nb_documents = index
+                .number_of_documents(&rtxn)
+                .map_err(|e| Error::from_milli(e, Some(uid.to_string())))?
+                as u32;
+            let (atomic, update_document_progress) = AtomicDocumentStep::new(nb_documents);
+            progress.update_progress(update_document_progress);
+            let documents = index
+                .all_documents(&rtxn)
+                .map_err(|e| Error::from_milli(e, Some(uid.to_string())))?;
+            // 4.1. Dump the documents
+            for ret in documents {
+                if self.scheduler.must_stop_processing.get() {
+                    return Err(Error::AbortedTask);
+                }
+
+                let (id, doc) = ret.map_err(|e| Error::from_milli(e, Some(uid.to_string())))?;
+
+                let mut document = milli::obkv_to_json(&all_fields, &fields_ids_map, doc)
+                    .map_err(|e| Error::from_milli(e, Some(uid.to_string())))?;
+
+                'inject_vectors: {
+                    let embeddings = index
+                        .embeddings(&rtxn, id)
+                        .map_err(|e| Error::from_milli(e, Some(uid.to_string())))?;
+
+                    if embeddings.is_empty() {
+                        break 'inject_vectors;
+                    }
+
+                    let vectors = document
+                        .entry(RESERVED_VECTORS_FIELD_NAME.to_owned())
+                        .or_insert(serde_json::Value::Object(Default::default()));
+
+                    let serde_json::Value::Object(vectors) = vectors else {
+                        let user_err =
+                            milli::Error::UserError(milli::UserError::InvalidVectorsMapType {
+                                document_id: {
+                                    if let Ok(Some(Ok(index))) = index
+                                        .external_id_of(&rtxn, std::iter::once(id))
+                                        .map(|it| it.into_iter().next())
+                                    {
+                                        index
+                                    } else {
+                                        format!("internal docid={id}")
+                                    }
+                                },
+                                value: vectors.clone(),
+                            });
+
+                        return Err(Error::from_milli(user_err, Some(uid.to_string())));
+                    };
+
+                    for (embedder_name, embeddings) in embeddings {
+                        let user_provided = embedding_configs
+                            .iter()
+                            .find(|conf| conf.name == embedder_name)
+                            .is_some_and(|conf| conf.user_provided.contains(id));
+                        let embeddings = ExplicitVectors {
+                            embeddings: Some(VectorOrArrayOfVectors::from_array_of_vectors(
+                                embeddings,
+                            )),
+                            regenerate: !user_provided,
+                        };
+                        vectors.insert(embedder_name, serde_json::to_value(embeddings).unwrap());
+                    }
+                }
+
+                index_dumper.push_document(&document)?;
+                atomic.fetch_add(1, Ordering::Relaxed);
+            }
+
+            // 4.2. Dump the settings
+            let settings = meilisearch_types::settings::settings(
+                index,
+                &rtxn,
+                meilisearch_types::settings::SecretPolicy::RevealSecrets,
+            )
+            .map_err(|e| Error::from_milli(e, Some(uid.to_string())))?;
+            index_dumper.settings(&settings)?;
+            Ok(())
+        })?;
+
+        // 5. Dump experimental feature settings
+        progress.update_progress(DumpCreationProgress::DumpTheExperimentalFeatures);
+        let features = self.features().runtime_features();
+        dump.create_experimental_features(features)?;
+        let network = self.network();
+        dump.create_network(network)?;
+
+        let dump_uid = started_at.format(format_description!(
+                    "[year repr:full][month repr:numerical][day padding:zero]-[hour padding:zero][minute padding:zero][second padding:zero][subsecond digits:3]"
+                )).unwrap();
+
+        if self.scheduler.must_stop_processing.get() {
+            return Err(Error::AbortedTask);
+        }
+        progress.update_progress(DumpCreationProgress::CompressTheDump);
+        let path = self.scheduler.dumps_path.join(format!("{}.dump", dump_uid));
+        let file = File::create(path)?;
+        dump.persist_to(BufWriter::new(file))?;
+
+        // if we reached this step we can tell the scheduler we succeeded to dump ourselves.
+        task.status = Status::Succeeded;
+        task.details = Some(Details::Dump { dump_uid: Some(dump_uid) });
+        Ok(vec![task])
+    }
+}