Bump version: 0.8.0-beta.0 → 0.8.0

The wheels are built to `WORKDIR/target/wheels` and the step was
configured to look for them at `WORKDIR/python/target/wheels`.

.bumpversion.cfg

@@ -1,22 +0,0 @@
-[bumpversion]
-current_version = 0.4.20
-commit = True
-message = Bump version: {current_version} → {new_version}
-tag = True
-tag_name = v{new_version}
-
-[bumpversion:file:node/package.json]
-
-[bumpversion:file:nodejs/package.json]
-
-[bumpversion:file:nodejs/npm/darwin-x64/package.json]
-
-[bumpversion:file:nodejs/npm/darwin-arm64/package.json]
-
-[bumpversion:file:nodejs/npm/linux-x64-gnu/package.json]
-
-[bumpversion:file:nodejs/npm/linux-arm64-gnu/package.json]
-
-[bumpversion:file:rust/ffi/node/Cargo.toml]
-
-[bumpversion:file:rust/lancedb/Cargo.toml]

.bumpversion.toml

@@ -0,0 +1,57 @@
+[tool.bumpversion]
+current_version = "0.5.0"
+parse = """(?x)
+    (?P<major>0|[1-9]\\d*)\\.
+    (?P<minor>0|[1-9]\\d*)\\.
+    (?P<patch>0|[1-9]\\d*)
+    (?:-(?P<pre_l>[a-zA-Z-]+)\\.(?P<pre_n>0|[1-9]\\d*))?
+"""
+serialize = [
+    "{major}.{minor}.{patch}-{pre_l}.{pre_n}",
+    "{major}.{minor}.{patch}",
+]
+search = "{current_version}"
+replace = "{new_version}"
+regex = false
+ignore_missing_version = false
+ignore_missing_files = false
+tag = true
+sign_tags = false
+tag_name = "v{new_version}"
+tag_message = "Bump version: {current_version} → {new_version}"
+allow_dirty = true
+commit = true
+message = "Bump version: {current_version} → {new_version}"
+commit_args = ""
+
+[tool.bumpversion.parts.pre_l]
+values = ["beta", "final"]
+optional_value = "final"
+
+[[tool.bumpversion.files]]
+filename = "node/package.json"
+search = "\"version\": \"{current_version}\","
+replace = "\"version\": \"{new_version}\","
+
+[[tool.bumpversion.files]]
+filename = "nodejs/package.json"
+search = "\"version\": \"{current_version}\","
+replace = "\"version\": \"{new_version}\","
+
+# nodejs binary packages
+[[tool.bumpversion.files]]
+glob = "nodejs/npm/*/package.json"
+search = "\"version\": \"{current_version}\","
+replace = "\"version\": \"{new_version}\","
+
+# Cargo files
+# ------------
+[[tool.bumpversion.files]]
+filename = "rust/ffi/node/Cargo.toml"
+search = "\nversion = \"{current_version}\""
+replace = "\nversion = \"{new_version}\""
+
+[[tool.bumpversion.files]]
+filename = "rust/lancedb/Cargo.toml"
+search = "\nversion = \"{current_version}\""
+replace = "\nversion = \"{new_version}\""

.github/release.yml

@@ -1,25 +0,0 @@
-# TODO: create separate templates for Python and other releases.
-changelog:
-  exclude:
-    labels:
-      - ci
-      - chore
-  categories:
-    - title: Breaking Changes 🛠
-      labels:
-        - breaking-change
-    - title: New Features 🎉
-      labels:
-        - enhancement
-    - title: Bug Fixes 🐛
-      labels:
-        - bug
-    - title: Documentation 📚
-      labels:
-        - documentation
-    - title: Performance Improvements 🚀
-      labels:
-        - performance
-    - title: Other Changes
-      labels:
-        - "*"

.github/release_notes.json

@@ -0,0 +1,41 @@
+{
+    "ignore_labels": ["chore"],
+    "pr_template": "- ${{TITLE}} by @${{AUTHOR}} in ${{URL}}",
+    "categories": [
+        {
+            "title": "## 🏆 Highlights",
+            "labels": ["highlight"]
+        },
+        {
+            "title": "## 🛠 Breaking Changes",
+            "labels": ["breaking-change"]
+        },
+        {
+            "title": "## ⚠️ Deprecations ",
+            "labels": ["deprecation"]
+        },
+        {
+            "title": "## 🎉 New Features",
+            "labels": ["enhancement"]
+        },
+        {
+            "title": "## 🐛 Bug Fixes",
+            "labels": ["bug"]
+        },
+        {
+            "title": "## 📚 Documentation",
+            "labels": ["documentation"]
+        },
+        {
+            "title": "## 🚀 Performance Improvements",
+            "labels": ["performance"]
+        },
+        {
+            "title": "## Other Changes"
+        },
+        {
+            "title": "## 🔧 Build and CI",
+            "labels": ["ci"]
+        }
+    ]
+}

.github/workflows/cargo-publish.yml

@@ -1,8 +1,12 @@
 name: Cargo Publish
 
 on:
-  release:
+  push:
-    types: [ published ]
+    tags-ignore:
+      # We don't publish pre-releases for Rust. Crates.io is just a source
+      # distribution, so we don't need to publish pre-releases.
+      - 'v*-beta*'
+      - '*-v*' # for example, python-vX.Y.Z
 
 env:
   # This env var is used by Swatinem/rust-cache@v2 for the cache

.github/workflows/make-release-commit.yml

@@ -1,37 +1,62 @@
 name: Create release commit
 
+# This workflow increments versions, tags the version, and pushes it.
+# When a tag is pushed, another workflow is triggered that creates a GH release
+# and uploads the binaries. This workflow is only for creating the tag.
+
+# This script will enforce that a minor version is incremented if there are any
+# breaking changes since the last minor increment. However, it isn't able to
+# differentiate between breaking changes in Node versus Python. If you wish to
+# bypass this check, you can manually increment the version and push the tag.
 on:
   workflow_dispatch:
     inputs:
       dry_run:
         description: 'Dry run (create the local commit/tags but do not push it)'
         required: true
-        default: "false"
+        default: false
-        type: choice
+        type: boolean
-        options:
+      type:
-          - "true"
-          - "false"
-      part:
         description: 'What kind of release is this?'
         required: true
-        default: 'patch'
+        default: 'preview'
         type: choice
         options:
-          - patch
+          - preview
-          - minor
+          - stable
-          - major
+      python:
+        description: 'Make a Python release'
+        required: true
+        default: true
+        type: boolean
+      other:
+        description: 'Make a Node/Rust release'
+        required: true
+        default: true
+        type: boolean
+      bump-minor:
+        description: 'Bump minor version'
+        required: true
+        default: false
+        type: boolean
 
 jobs:
-  bump-version:
+  make-release:
+    # Creates tag and GH release. The GH release will trigger the build and release jobs.
     runs-on: ubuntu-latest
+    permissions:
+      contents: write
     steps:
-      - name: Check out main
+      - name: Output Inputs
-        uses: actions/checkout@v4
+        run: echo "${{ toJSON(github.event.inputs) }}"
+      - uses: actions/checkout@v4
         with:
-          ref: main
-          persist-credentials: false
           fetch-depth: 0
           lfs: true
+          # It's important we use our token here, as the default token will NOT
+          # trigger any workflows watching for new tags. See:
+          # https://docs.github.com/en/actions/using-workflows/triggering-a-workflow#triggering-a-workflow-from-a-workflow
+          token: ${{ secrets.LANCEDB_RELEASE_TOKEN }}
       - name: Set git configs for bumpversion
         shell: bash
         run: |
@@ -41,19 +66,34 @@ jobs:
         uses: actions/setup-python@v5
         with:
           python-version: "3.11"
-      - name: Bump version, create tag and commit
+      - name: Bump Python version
+        if: ${{ inputs.python }}
+        working-directory: python
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
         run: |
-          pip install bump2version
+          # Need to get the commit before bumping the version, so we can
-          bumpversion --verbose ${{ inputs.part }}
+          # determine if there are breaking changes in the next step as well.
-      - name: Push new version and tag
+          echo "COMMIT_BEFORE_BUMP=$(git rev-parse HEAD)" >> $GITHUB_ENV
-        if: ${{ inputs.dry_run }} == "false"
+
+          pip install bump-my-version PyGithub packaging
+          bash ../ci/bump_version.sh ${{ inputs.type }} ${{ inputs.bump-minor }} python-v
+      - name: Bump Node/Rust version
+        if: ${{ inputs.other }}
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        run: |
+          pip install bump-my-version PyGithub packaging
+          bash ci/bump_version.sh ${{ inputs.type }} ${{ inputs.bump-minor }} v $COMMIT_BEFORE_BUMP
+      - name: Push new version tag
+        if: ${{ !inputs.dry_run }}
         uses: ad-m/github-push-action@master
         with:
+          # Need to use PAT here too to trigger next workflow. See comment above.
           github_token: ${{ secrets.LANCEDB_RELEASE_TOKEN }}
-          branch: main
+          branch: ${{ github.ref }}
           tags: true
       - uses: ./.github/workflows/update_package_lock
         if: ${{ inputs.dry_run }} == "false"
         with:
-          github_token: ${{ secrets.LANCEDB_RELEASE_TOKEN }}
+          github_token: ${{ secrets.GITHUB_TOKEN }}
-

.github/workflows/npm-publish.yml

@@ -1,8 +1,9 @@
 name: NPM Publish
 
 on:
-  release:
+  push:
-    types: [published]
+    tags:
+      - 'v*'
 
 jobs:
   node:
@@ -274,9 +275,15 @@ jobs:
         env:
           NODE_AUTH_TOKEN: ${{ secrets.LANCEDB_NPM_REGISTRY_TOKEN }}
         run: |
+          # Tag beta as "preview" instead of default "latest". See lancedb 
+          # npm publish step for more info.
+          if [[ $GITHUB_REF =~ refs/tags/v(.*)-beta.* ]]; then
+            PUBLISH_ARGS="--tag preview"
+          fi
+
           mv */*.tgz .
           for filename in *.tgz; do
-            npm publish $filename
+            npm publish $PUBLISH_ARGS $filename
           done
 
   release-nodejs:
@@ -316,11 +323,23 @@ jobs:
       - name: Publish to NPM
         env:
           NODE_AUTH_TOKEN: ${{ secrets.LANCEDB_NPM_REGISTRY_TOKEN }}
-        run: npm publish --access public
+        # By default, things are published to the latest tag. This is what is 
+        # installed by default if the user does not specify a version. This is
+        # good for stable releases, but for pre-releases, we want to publish to
+        # the "preview" tag so they can install with `npm install lancedb@preview`.
+        # See: https://medium.com/@mbostock/prereleases-and-npm-e778fc5e2420
+        run: |
+          if [[ $GITHUB_REF =~ refs/tags/v(.*)-beta.* ]]; then
+            npm publish --access public --tag preview
+          else
+            npm publish --access public
+          fi
 
   update-package-lock:
     needs: [release]
     runs-on: ubuntu-latest
+    permissions:
+      contents: write
     steps:
       - name: Checkout
         uses: actions/checkout@v4
@@ -331,11 +350,13 @@ jobs:
           lfs: true
       - uses: ./.github/workflows/update_package_lock
         with:
-          github_token: ${{ secrets.LANCEDB_RELEASE_TOKEN }}
+          github_token: ${{ secrets.GITHUB_TOKEN }}
 
   update-package-lock-nodejs:
     needs: [release-nodejs]
     runs-on: ubuntu-latest
+    permissions:
+      contents: write
     steps:
       - name: Checkout
         uses: actions/checkout@v4
@@ -346,4 +367,70 @@ jobs:
           lfs: true
       - uses: ./.github/workflows/update_package_lock_nodejs
         with:
-          github_token: ${{ secrets.LANCEDB_RELEASE_TOKEN }}
+          github_token: ${{ secrets.GITHUB_TOKEN }}
+  
+  gh-release:
+    runs-on: ubuntu-latest
+    permissions:
+      contents: write
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          lfs: true
+      - name: Extract version
+        id: extract_version
+        env:
+          GITHUB_REF: ${{ github.ref }}
+        run: |
+          set -e
+          echo "Extracting tag and version from $GITHUB_REF"
+          if [[ $GITHUB_REF =~ refs/tags/v(.*) ]]; then
+            VERSION=${BASH_REMATCH[1]}
+            TAG=v$VERSION
+            echo "tag=$TAG" >> $GITHUB_OUTPUT
+            echo "version=$VERSION" >> $GITHUB_OUTPUT
+          else
+            echo "Failed to extract version from $GITHUB_REF"
+            exit 1
+          fi
+          echo "Extracted version $VERSION from $GITHUB_REF"
+          if [[ $VERSION =~ beta ]]; then
+            echo "This is a beta release"
+
+            # Get last release (that is not this one)
+            FROM_TAG=$(git tag --sort='version:refname' \
+              | grep ^v \
+              | grep -vF "$TAG" \
+              | python ci/semver_sort.py v \
+              | tail -n 1)
+          else
+            echo "This is a stable release"
+            # Get last stable tag (ignore betas)
+            FROM_TAG=$(git tag --sort='version:refname' \
+              | grep ^v \
+              | grep -vF "$TAG" \
+              | grep -v beta \
+              | python ci/semver_sort.py v \
+              | tail -n 1)
+          fi
+          echo "Found from tag $FROM_TAG"
+          echo "from_tag=$FROM_TAG" >> $GITHUB_OUTPUT
+      - name: Create Release Notes
+        id: release_notes
+        uses: mikepenz/release-changelog-builder-action@v4
+        with:
+          configuration: .github/release_notes.json
+          toTag: ${{ steps.extract_version.outputs.tag }}
+          fromTag: ${{ steps.extract_version.outputs.from_tag }}
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+      - name: Create GH release
+        uses: softprops/action-gh-release@v2
+        with:
+          prerelease: ${{ contains('beta', github.ref) }}
+          tag_name: ${{ steps.extract_version.outputs.tag }}
+          token: ${{ secrets.GITHUB_TOKEN }}
+          generate_release_notes: false
+          name: Node/Rust LanceDB v${{ steps.extract_version.outputs.version }}
+          body: ${{ steps.release_notes.outputs.changelog }}

.github/workflows/pypi-publish.yml

@@ -1,18 +1,16 @@
 name: PyPI Publish
 
 on:
-  release:
+  push:
-    types: [published]
+    tags:
+      - 'python-v*'
 
 jobs:
   linux:
-    # Only runs on tags that matches the python-make-release action
-    if: startsWith(github.ref, 'refs/tags/python-v')
     name: Python ${{ matrix.config.platform }} manylinux${{ matrix.config.manylinux }}
     timeout-minutes: 60
     strategy:
       matrix:
-        python-minor-version: ["8"]
         config:
           - platform: x86_64
             manylinux: "2_17"
@@ -34,25 +32,22 @@ jobs:
       - name: Set up Python
         uses: actions/setup-python@v4
         with:
-          python-version: 3.${{ matrix.python-minor-version }}
+          python-version: 3.8
       - uses: ./.github/workflows/build_linux_wheel
         with:
-          python-minor-version: ${{ matrix.python-minor-version }}
+          python-minor-version: 8
           args: "--release --strip ${{ matrix.config.extra_args }}"
           arm-build: ${{ matrix.config.platform == 'aarch64' }}
           manylinux: ${{ matrix.config.manylinux }}
       - uses: ./.github/workflows/upload_wheel
         with:
-          token: ${{ secrets.LANCEDB_PYPI_API_TOKEN }}
+          pypi_token: ${{ secrets.LANCEDB_PYPI_API_TOKEN }}
-          repo: "pypi"
+          fury_token: ${{ secrets.FURY_TOKEN }}
   mac:
-    # Only runs on tags that matches the python-make-release action
-    if: startsWith(github.ref, 'refs/tags/python-v')
     timeout-minutes: 60
     runs-on: ${{ matrix.config.runner }}
     strategy:
       matrix:
-        python-minor-version: ["8"]
         config:
           - target: x86_64-apple-darwin
             runner: macos-13
@@ -63,7 +58,6 @@ jobs:
     steps:
       - uses: actions/checkout@v4
         with:
-          ref: ${{ inputs.ref }}
           fetch-depth: 0
           lfs: true
       - name: Set up Python
@@ -72,38 +66,95 @@ jobs:
           python-version: 3.12
       - uses: ./.github/workflows/build_mac_wheel
         with:
-          python-minor-version: ${{ matrix.python-minor-version }}
+          python-minor-version: 8
           args: "--release --strip --target ${{ matrix.config.target }} --features fp16kernels"
       - uses: ./.github/workflows/upload_wheel
         with:
-          python-minor-version: ${{ matrix.python-minor-version }}
+          pypi_token: ${{ secrets.LANCEDB_PYPI_API_TOKEN }}
-          token: ${{ secrets.LANCEDB_PYPI_API_TOKEN }}
+          fury_token: ${{ secrets.FURY_TOKEN }}
-          repo: "pypi"
   windows:
-    # Only runs on tags that matches the python-make-release action
-    if: startsWith(github.ref, 'refs/tags/python-v')
     timeout-minutes: 60
     runs-on: windows-latest
-    strategy:
-      matrix:
-        python-minor-version: ["8"]
     steps:
       - uses: actions/checkout@v4
         with:
-          ref: ${{ inputs.ref }}
           fetch-depth: 0
           lfs: true
       - name: Set up Python
         uses: actions/setup-python@v4
         with:
-          python-version: 3.${{ matrix.python-minor-version }}
+          python-version: 3.8
       - uses: ./.github/workflows/build_windows_wheel
         with:
-          python-minor-version: ${{ matrix.python-minor-version }}
+          python-minor-version: 8
           args: "--release --strip"
           vcpkg_token: ${{ secrets.VCPKG_GITHUB_PACKAGES }}
       - uses: ./.github/workflows/upload_wheel
         with:
-          python-minor-version: ${{ matrix.python-minor-version }}
+          pypi_token: ${{ secrets.LANCEDB_PYPI_API_TOKEN }}
-          token: ${{ secrets.LANCEDB_PYPI_API_TOKEN }}
+          fury_token: ${{ secrets.FURY_TOKEN }}
-          repo: "pypi"
+  gh-release:
+    runs-on: ubuntu-latest
+    permissions:
+      contents: write
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          lfs: true
+      - name: Extract version
+        id: extract_version
+        env:
+          GITHUB_REF: ${{ github.ref }}
+        run: |
+          set -e
+          echo "Extracting tag and version from $GITHUB_REF"
+          if [[ $GITHUB_REF =~ refs/tags/python-v(.*) ]]; then
+            VERSION=${BASH_REMATCH[1]}
+            TAG=python-v$VERSION
+            echo "tag=$TAG" >> $GITHUB_OUTPUT
+            echo "version=$VERSION" >> $GITHUB_OUTPUT
+          else
+            echo "Failed to extract version from $GITHUB_REF"
+            exit 1
+          fi
+          echo "Extracted version $VERSION from $GITHUB_REF"
+          if [[ $VERSION =~ beta ]]; then
+            echo "This is a beta release"
+
+            # Get last release (that is not this one)
+            FROM_TAG=$(git tag --sort='version:refname' \
+              | grep ^python-v \
+              | grep -vF "$TAG" \
+              | python ci/semver_sort.py python-v \
+              | tail -n 1)
+          else
+            echo "This is a stable release"
+            # Get last stable tag (ignore betas)
+            FROM_TAG=$(git tag --sort='version:refname' \
+              | grep ^python-v \
+              | grep -vF "$TAG" \
+              | grep -v beta \
+              | python ci/semver_sort.py python-v \
+              | tail -n 1)
+          fi
+          echo "Found from tag $FROM_TAG"
+          echo "from_tag=$FROM_TAG" >> $GITHUB_OUTPUT
+      - name: Create Python Release Notes
+        id: python_release_notes
+        uses: mikepenz/release-changelog-builder-action@v4
+        with:
+          configuration: .github/release_notes.json
+          toTag: ${{ steps.extract_version.outputs.tag }}
+          fromTag: ${{ steps.extract_version.outputs.from_tag }}
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+      - name: Create Python GH release
+        uses: softprops/action-gh-release@v2
+        with:
+          prerelease: ${{ contains('beta', github.ref) }}
+          tag_name: ${{ steps.extract_version.outputs.tag }}
+          token: ${{ secrets.GITHUB_TOKEN }}
+          generate_release_notes: false
+          name: Python LanceDB v${{ steps.extract_version.outputs.version }}
+          body: ${{ steps.python_release_notes.outputs.changelog }}

.github/workflows/python-make-release-commit.yml

@@ -1,56 +0,0 @@
-name: Python - Create release commit
-
-on:
-  workflow_dispatch:
-    inputs:
-      dry_run:
-        description: 'Dry run (create the local commit/tags but do not push it)'
-        required: true
-        default: "false"
-        type: choice
-        options:
-          - "true"
-          - "false"
-      part:
-        description: 'What kind of release is this?'
-        required: true
-        default: 'patch'
-        type: choice
-        options:
-          - patch
-          - minor
-          - major
-
-jobs:
-  bump-version:
-    runs-on: ubuntu-latest
-    steps:
-    - name: Check out main
-      uses: actions/checkout@v4
-      with:
-        ref: main
-        persist-credentials: false
-        fetch-depth: 0
-        lfs: true
-    - name: Set git configs for bumpversion
-      shell: bash
-      run: |
-        git config user.name 'Lance Release'
-        git config user.email 'lance-dev@lancedb.com'
-    - name: Set up Python
-      uses: actions/setup-python@v5
-      with:
-        python-version: "3.11"
-    - name: Bump version, create tag and commit
-      working-directory: python
-      run: |
-        pip install bump2version
-        bumpversion --verbose ${{ inputs.part }}
-    - name: Push new version and tag
-      if: ${{ inputs.dry_run }} == "false"
-      uses: ad-m/github-push-action@master
-      with:
-        github_token: ${{ secrets.LANCEDB_RELEASE_TOKEN }}
-        branch: main
-        tags: true
-

.github/workflows/python.yml

@@ -75,7 +75,7 @@ jobs:
     timeout-minutes: 30
     strategy:
       matrix:
-        python-minor-version: ["8", "11"]
+        python-minor-version: ["9", "11"]
     runs-on: "ubuntu-22.04"
     defaults:
       run:

.github/workflows/rust.yml

@@ -74,11 +74,11 @@ jobs:
       run: |
           sudo apt update
           sudo apt install -y protobuf-compiler libssl-dev
-    - name: Build
-      run: cargo build --all-features
     - name: Start S3 integration test environment
       working-directory: .
       run: docker compose up --detach --wait
+    - name: Build
+      run: cargo build --all-features
     - name: Run tests
       run: cargo test --all-features
     - name: Run examples

.github/workflows/upload_wheel/action.yml

@@ -2,28 +2,43 @@ name: upload-wheel
 
 description: "Upload wheels to Pypi"
 inputs:
-  os:
+  pypi_token:
-    required: true
-    description: "ubuntu-22.04 or macos-13"
-  repo:
-    required: false
-    description: "pypi or testpypi"
-    default: "pypi"
-  token:
     required: true
     description: "release token for the repo"
+  fury_token:
+    required: true
+    description: "release token for the fury repo"
 
 runs:
   using: "composite"
   steps:
-    - name: Install dependencies
+  - name: Install dependencies
-      shell: bash
+    shell: bash
-      run: |
+    run: |
-        python -m pip install --upgrade pip
+      python -m pip install --upgrade pip
-        pip install twine
+      pip install twine
-    - name: Publish wheel
+  - name: Choose repo
-      env:
+    shell: bash
-        TWINE_USERNAME: __token__
+    id: choose_repo
-        TWINE_PASSWORD: ${{ inputs.token }}
+    run: |
-      shell: bash
+      if [ ${{ github.ref }} == "*beta*" ]; then
-      run: twine upload --repository ${{ inputs.repo }} target/wheels/lancedb-*.whl
+        echo "repo=fury" >> $GITHUB_OUTPUT
+      else
+        echo "repo=pypi" >> $GITHUB_OUTPUT
+      fi
+  - name: Publish to PyPI
+    shell: bash
+    env:
+      FURY_TOKEN: ${{ inputs.fury_token }}
+      PYPI_TOKEN: ${{ inputs.pypi_token }}
+    run: |
+      if [ ${{ steps.choose_repo.outputs.repo }} == "fury" ]; then
+        WHEEL=$(ls target/wheels/lancedb-*.whl 2> /dev/null | head -n 1)
+        echo "Uploading $WHEEL to Fury"
+        curl -f -F package=@$WHEEL https://$FURY_TOKEN@push.fury.io/lancedb/
+      else
+        twine upload --repository ${{ steps.choose_repo.outputs.repo }} \
+          --username __token__ \
+          --password $PYPI_TOKEN \
+          target/wheels/lancedb-*.whl
+      fi

Cargo.toml

@@ -14,10 +14,10 @@ keywords = ["lancedb", "lance", "database", "vector", "search"]
 categories = ["database-implementations"]
 
 [workspace.dependencies]
-lance = { "version" = "=0.10.18", "features" = ["dynamodb"] }
+lance = { "version" = "=0.11.0", "features" = ["dynamodb"] }
-lance-index = { "version" = "=0.10.18" }
+lance-index = { "version" = "=0.11.0" }
-lance-linalg = { "version" = "=0.10.18" }
+lance-linalg = { "version" = "=0.11.0" }
-lance-testing = { "version" = "=0.10.18" }
+lance-testing = { "version" = "=0.11.0" }
 # Note that this one does not include pyarrow
 arrow = { version = "51.0", optional = false }
 arrow-array = "51.0"

ci/bump_version.sh

@@ -0,0 +1,51 @@
+set -e
+
+RELEASE_TYPE=${1:-"stable"}
+BUMP_MINOR=${2:-false}
+TAG_PREFIX=${3:-"v"} # Such as "python-v"
+HEAD_SHA=${4:-$(git rev-parse HEAD)}
+
+readonly SELF_DIR=$(cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )
+
+PREV_TAG=$(git tag --sort='version:refname' | grep ^$TAG_PREFIX | python $SELF_DIR/semver_sort.py $TAG_PREFIX | tail -n 1)
+echo "Found previous tag $PREV_TAG"
+
+# Initially, we don't want to tag if we are doing stable, because we will bump
+# again later. See comment at end for why.
+if [[ "$RELEASE_TYPE" == 'stable' ]]; then 
+  BUMP_ARGS="--no-tag"
+fi
+
+# If last is stable and not bumping minor
+if [[ $PREV_TAG != *beta* ]]; then
+    if [[ "$BUMP_MINOR" != "false" ]]; then
+      # X.Y.Z -> X.(Y+1).0-beta.0
+      bump-my-version bump -vv $BUMP_ARGS minor
+    else
+      # X.Y.Z -> X.Y.(Z+1)-beta.0
+      bump-my-version bump -vv $BUMP_ARGS patch
+    fi
+else
+  if [[ "$BUMP_MINOR" != "false" ]]; then
+    # X.Y.Z-beta.N -> X.(Y+1).0-beta.0
+    bump-my-version bump -vv $BUMP_ARGS minor
+  else
+    # X.Y.Z-beta.N -> X.Y.Z-beta.(N+1)
+    bump-my-version bump -vv $BUMP_ARGS pre_n
+  fi
+fi
+
+# The above bump will always bump to a pre-release version. If we are releasing
+# a stable version, bump the pre-release level ("pre_l") to make it stable.
+if [[ $RELEASE_TYPE == 'stable' ]]; then
+  # X.Y.Z-beta.N -> X.Y.Z
+  bump-my-version bump -vv pre_l
+fi
+
+# Validate that we have incremented version appropriately for breaking changes
+NEW_TAG=$(git describe --tags --exact-match HEAD)
+NEW_VERSION=$(echo $NEW_TAG | sed "s/^$TAG_PREFIX//")
+LAST_STABLE_RELEASE=$(git tag --sort='version:refname' | grep ^$TAG_PREFIX | grep -v beta | grep -vF "$NEW_TAG" | python $SELF_DIR/semver_sort.py $TAG_PREFIX | tail -n 1)
+LAST_STABLE_VERSION=$(echo $LAST_STABLE_RELEASE | sed "s/^$TAG_PREFIX//")
+
+python $SELF_DIR/check_breaking_changes.py $LAST_STABLE_RELEASE $HEAD_SHA $LAST_STABLE_VERSION $NEW_VERSION

ci/check_breaking_changes.py

@@ -0,0 +1,35 @@
+"""
+Check whether there are any breaking changes in the PRs between the base and head commits.
+If there are, assert that we have incremented the minor version.
+"""
+import argparse
+import os
+from packaging.version import parse
+
+from github import Github
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("base")
+    parser.add_argument("head")
+    parser.add_argument("last_stable_version")
+    parser.add_argument("current_version")
+    args = parser.parse_args()
+
+    repo = Github(os.environ["GITHUB_TOKEN"]).get_repo(os.environ["GITHUB_REPOSITORY"])
+    commits = repo.compare(args.base, args.head).commits
+    prs = (pr for commit in commits for pr in commit.get_pulls())
+
+    for pr in prs:
+        if any(label.name == "breaking-change" for label in pr.labels):
+            print(f"Breaking change in PR: {pr.html_url}")
+            break
+    else:
+        print("No breaking changes found.")
+        exit(0)
+    
+    last_stable_version = parse(args.last_stable_version)
+    current_version = parse(args.current_version)
+    if current_version.minor <= last_stable_version.minor:
+        print("Minor version is not greater than the last stable version.")
+        exit(1)

ci/semver_sort.py

@@ -0,0 +1,35 @@
+"""
+Takes a list of semver strings and sorts them in ascending order.
+"""
+
+import sys
+from packaging.version import parse, InvalidVersion
+
+if __name__ == "__main__":
+    import argparse
+    parser = argparse.ArgumentParser()
+    parser.add_argument("prefix", default="v")
+    args = parser.parse_args()
+
+    # Read the input from stdin
+    lines = sys.stdin.readlines()
+
+    # Parse the versions
+    versions = []
+    for line in lines:
+        line = line.strip()
+        try:
+            version_str = line.removeprefix(args.prefix)
+            version = parse(version_str)
+        except InvalidVersion:
+            # There are old tags that don't follow the semver format
+            print(f"Invalid version: {line}", file=sys.stderr)
+            continue
+        versions.append((line, version))
+
+    # Sort the versions
+    versions.sort(key=lambda x: x[1])
+
+    # Print the sorted versions as original strings
+    for line, _ in versions:
+        print(line)

docs/src/basic.md

@@ -44,6 +44,36 @@
 
     !!! info "Please also make sure you're using the same version of Arrow as in the [lancedb crate](https://github.com/lancedb/lancedb/blob/main/Cargo.toml)"
 
+### Preview releases
+
+Stable releases are created about every 2 weeks. For the latest features and bug
+fixes, you can install the preview release. These releases receive the same
+level of testing as stable releases, but are not guaranteed to be available for
+more than 6 months after they are released. Once your application is stable, we
+recommend switching to stable releases.
+
+=== "Python"
+
+      ```shell
+      pip install --pre --extra-index-url https://pypi.fury.io/lancedb/ lancedb
+      ```
+
+=== "Typescript"
+
+      ```shell
+      npm install vectordb@preview
+      ```
+
+=== "Rust"
+    
+    We don't push preview releases to crates.io, but you can referent the tag
+    in GitHub within your Cargo dependencies:
+
+    ```toml
+    [dependencies]
+    lancedb = { git = "https://github.com/lancedb/lancedb.git", tag = "vX.Y.Z-beta.N" }
+    ```
+
 ## Connect to a database
 
 === "Python"

node/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "vectordb",
-  "version": "0.4.20",
+  "version": "0.5.0",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "vectordb",
-      "version": "0.4.20",
+      "version": "0.5.0",
       "cpu": [
         "x64",
         "arm64"

node/package.json

@@ -1,6 +1,6 @@
 {
   "name": "vectordb",
-  "version": "0.4.20",
+  "version": "0.5.0",
   "description": " Serverless, low-latency vector database for AI applications",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",

node/src/arrow.ts

@@ -624,8 +624,6 @@ function validateSchemaEmbeddings(
   }
 
   if (missingEmbeddingFields.length > 0 && embeddings === undefined) {
-    console.log({ missingEmbeddingFields, embeddings });
-
     throw new Error(
       `Table has embeddings: "${missingEmbeddingFields
         .map((f) => f.name)
@@ -633,5 +631,5 @@ function validateSchemaEmbeddings(
     );
   }
 
-  return new Schema(fields);
+  return new Schema(fields, schema.metadata);
 }

nodejs/__test__/table.test.ts

@@ -419,3 +419,31 @@ describe("when dealing with versioning", () => {
     );
   });
 });
+
+describe("when optimizing a dataset", () => {
+  let tmpDir: tmp.DirResult;
+  let table: Table;
+  beforeEach(async () => {
+    tmpDir = tmp.dirSync({ unsafeCleanup: true });
+    const con = await connect(tmpDir.name);
+    table = await con.createTable("vectors", [{ id: 1 }]);
+    await table.add([{ id: 2 }]);
+  });
+  afterEach(() => {
+    tmpDir.removeCallback();
+  });
+
+  it("compacts files", async () => {
+    const stats = await table.optimize();
+    expect(stats.compaction.filesAdded).toBe(1);
+    expect(stats.compaction.filesRemoved).toBe(2);
+    expect(stats.compaction.fragmentsAdded).toBe(1);
+    expect(stats.compaction.fragmentsRemoved).toBe(2);
+  });
+
+  it("cleanups old versions", async () => {
+    const stats = await table.optimize({ cleanupOlderThan: new Date() });
+    expect(stats.prune.bytesRemoved).toBeGreaterThan(0);
+    expect(stats.prune.oldVersionsRemoved).toBe(3);
+  });
+});

nodejs/lancedb/arrow.ts

@@ -677,8 +677,6 @@ function validateSchemaEmbeddings(
   }
 
   if (missingEmbeddingFields.length > 0 && embeddings === undefined) {
-    console.log({ missingEmbeddingFields, embeddings });
-
     throw new Error(
       `Table has embeddings: "${missingEmbeddingFields
         .map((f) => f.name)
@@ -686,5 +684,5 @@ function validateSchemaEmbeddings(
     );
   }
 
-  return new Schema(fields);
+  return new Schema(fields, schema.metadata);
 }

nodejs/lancedb/table.ts

@@ -19,6 +19,7 @@ import {
   AddColumnsSql,
   ColumnAlteration,
   IndexConfig,
+  OptimizeStats,
   Table as _NativeTable,
 } from "./native";
 import { Query, VectorQuery } from "./query";
@@ -50,6 +51,23 @@ export interface UpdateOptions {
   where: string;
 }
 
+export interface OptimizeOptions {
+  /**
+   * If set then all versions older than the given date
+   * be removed.  The current version will never be removed.
+   * The default is 7 days
+   * @example
+   * // Delete all versions older than 1 day
+   * const olderThan = new Date();
+   * olderThan.setDate(olderThan.getDate() - 1));
+   * tbl.cleanupOlderVersions(olderThan);
+   *
+   * // Delete all versions except the current version
+   * tbl.cleanupOlderVersions(new Date());
+   */
+  cleanupOlderThan: Date;
+}
+
 /**
  * A Table is a collection of Records in a LanceDB Database.
  *
@@ -352,6 +370,48 @@ export class Table {
     await this.inner.restore();
   }
 
+  /**
+   * Optimize the on-disk data and indices for better performance.
+   *
+   * Modeled after ``VACUUM`` in PostgreSQL.
+   *
+   *  Optimization covers three operations:
+   *
+   *  - Compaction: Merges small files into larger ones
+   *  - Prune: Removes old versions of the dataset
+   *  - Index: Optimizes the indices, adding new data to existing indices
+   *
+   *
+   *  Experimental API
+   *  ----------------
+   *
+   *  The optimization process is undergoing active development and may change.
+   *  Our goal with these changes is to improve the performance of optimization and
+   *  reduce the complexity.
+   *
+   *  That being said, it is essential today to run optimize if you want the best
+   *  performance.  It should be stable and safe to use in production, but it our
+   *  hope that the API may be simplified (or not even need to be called) in the
+   *  future.
+   *
+   *  The frequency an application shoudl call optimize is based on the frequency of
+   *  data modifications.  If data is frequently added, deleted, or updated then
+   *  optimize should be run frequently.  A good rule of thumb is to run optimize if
+   *  you have added or modified 100,000 or more records or run more than 20 data
+   *  modification operations.
+   */
+  async optimize(options?: Partial<OptimizeOptions>): Promise<OptimizeStats> {
+    let cleanupOlderThanMs;
+    if (
+      options?.cleanupOlderThan !== undefined &&
+      options?.cleanupOlderThan !== null
+    ) {
+      cleanupOlderThanMs =
+        new Date().getTime() - options.cleanupOlderThan.getTime();
+    }
+    return await this.inner.optimize(cleanupOlderThanMs);
+  }
+
   /** List all indices that have been created with {@link Table.createIndex} */
   async listIndices(): Promise<IndexConfig[]> {
     return await this.inner.listIndices();

nodejs/npm/darwin-arm64/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-darwin-arm64",
-	"version": "0.4.20",
+	"version": "0.5.0",
 	"os": ["darwin"],
 	"cpu": ["arm64"],
 	"main": "lancedb.darwin-arm64.node",

nodejs/npm/darwin-x64/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-darwin-x64",
-	"version": "0.4.20",
+	"version": "0.5.0",
 	"os": ["darwin"],
 	"cpu": ["x64"],
 	"main": "lancedb.darwin-x64.node",

nodejs/npm/linux-arm64-gnu/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-arm64-gnu",
-	"version": "0.4.20",
+	"version": "0.5.0",
 	"os": ["linux"],
 	"cpu": ["arm64"],
 	"main": "lancedb.linux-arm64-gnu.node",

nodejs/npm/linux-x64-gnu/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-x64-gnu",
-	"version": "0.4.20",
+	"version": "0.5.0",
 	"os": ["linux"],
 	"cpu": ["x64"],
 	"main": "lancedb.linux-x64-gnu.node",

nodejs/npm/win32-x64-msvc/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-win32-x64-msvc",
-	"version": "0.4.14",
+	"version": "0.5.0",
 	"os": ["win32"],
 	"cpu": ["x64"],
 	"main": "lancedb.win32-x64-msvc.node",

nodejs/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lancedb/lancedb",
-  "version": "0.4.20",
+  "version": "0.5.0",
   "main": "./dist/index.js",
   "types": "./dist/index.d.ts",
   "napi": {

nodejs/src/table.rs

@@ -15,8 +15,8 @@
 use arrow_ipc::writer::FileWriter;
 use lancedb::ipc::ipc_file_to_batches;
 use lancedb::table::{
-    AddDataMode, ColumnAlteration as LanceColumnAlteration, NewColumnTransform,
+    AddDataMode, ColumnAlteration as LanceColumnAlteration, Duration, NewColumnTransform,
-    Table as LanceDbTable,
+    OptimizeAction, OptimizeOptions, Table as LanceDbTable,
 };
 use napi::bindgen_prelude::*;
 use napi_derive::napi;
@@ -263,6 +263,60 @@ impl Table {
         self.inner_ref()?.restore().await.default_error()
     }
 
+    #[napi]
+    pub async fn optimize(&self, older_than_ms: Option<i64>) -> napi::Result<OptimizeStats> {
+        let inner = self.inner_ref()?;
+
+        let older_than = if let Some(ms) = older_than_ms {
+            if ms == i64::MIN {
+                return Err(napi::Error::from_reason(format!(
+                    "older_than_ms can not be {}",
+                    i32::MIN,
+                )));
+            }
+            Duration::try_milliseconds(ms)
+        } else {
+            None
+        };
+
+        let compaction_stats = inner
+            .optimize(OptimizeAction::Compact {
+                options: lancedb::table::CompactionOptions::default(),
+                remap_options: None,
+            })
+            .await
+            .default_error()?
+            .compaction
+            .unwrap();
+        let prune_stats = inner
+            .optimize(OptimizeAction::Prune {
+                older_than,
+                delete_unverified: None,
+            })
+            .await
+            .default_error()?
+            .prune
+            .unwrap();
+        inner
+            .optimize(lancedb::table::OptimizeAction::Index(
+                OptimizeOptions::default(),
+            ))
+            .await
+            .default_error()?;
+        Ok(OptimizeStats {
+            compaction: CompactionStats {
+                files_added: compaction_stats.files_added as i64,
+                files_removed: compaction_stats.files_removed as i64,
+                fragments_added: compaction_stats.fragments_added as i64,
+                fragments_removed: compaction_stats.fragments_removed as i64,
+            },
+            prune: RemovalStats {
+                bytes_removed: prune_stats.bytes_removed as i64,
+                old_versions_removed: prune_stats.old_versions as i64,
+            },
+        })
+    }
+
     #[napi]
     pub async fn list_indices(&self) -> napi::Result<Vec<IndexConfig>> {
         Ok(self
@@ -298,6 +352,40 @@ impl From<lancedb::index::IndexConfig> for IndexConfig {
     }
 }
 
+/// Statistics about a compaction operation.
+#[napi(object)]
+#[derive(Clone, Debug)]
+pub struct CompactionStats {
+    /// The number of fragments removed
+    pub fragments_removed: i64,
+    /// The number of new, compacted fragments added
+    pub fragments_added: i64,
+    /// The number of data files removed
+    pub files_removed: i64,
+    /// The number of new, compacted data files added
+    pub files_added: i64,
+}
+
+/// Statistics about a cleanup operation
+#[napi(object)]
+#[derive(Clone, Debug)]
+pub struct RemovalStats {
+    /// The number of bytes removed
+    pub bytes_removed: i64,
+    /// The number of old versions removed
+    pub old_versions_removed: i64,
+}
+
+/// Statistics about an optimize operation
+#[napi(object)]
+#[derive(Clone, Debug)]
+pub struct OptimizeStats {
+    /// Statistics about the compaction operation
+    pub compaction: CompactionStats,
+    /// Statistics about the removal operation
+    pub prune: RemovalStats,
+}
+
 ///  A definition of a column alteration. The alteration changes the column at
 /// `path` to have the new name `name`, to be nullable if `nullable` is true,
 /// and to have the data type `data_type`. At least one of `rename` or `nullable`

python/.bumpversion.cfg

@@ -1,8 +0,0 @@
-[bumpversion]
-current_version = 0.6.13
-commit = True
-message = [python] Bump version: {current_version} → {new_version}
-tag = True
-tag_name = python-v{new_version}
-
-[bumpversion:file:pyproject.toml]

python/.bumpversion.toml

@@ -0,0 +1,34 @@
+[tool.bumpversion]
+current_version = "0.8.0"
+parse = """(?x)
+    (?P<major>0|[1-9]\\d*)\\.
+    (?P<minor>0|[1-9]\\d*)\\.
+    (?P<patch>0|[1-9]\\d*)
+    (?:-(?P<pre_l>[a-zA-Z-]+)\\.(?P<pre_n>0|[1-9]\\d*))?
+"""
+serialize = [
+    "{major}.{minor}.{patch}-{pre_l}.{pre_n}",
+    "{major}.{minor}.{patch}",
+]
+search = "{current_version}"
+replace = "{new_version}"
+regex = false
+ignore_missing_version = false
+ignore_missing_files = false
+tag = true
+sign_tags = false
+tag_name = "python-v{new_version}"
+tag_message = "Bump version: {current_version} → {new_version}"
+allow_dirty = true
+commit = true
+message = "Bump version: {current_version} → {new_version}"
+commit_args = ""
+
+[tool.bumpversion.parts.pre_l]
+values = ["beta", "final"]
+optional_value = "final"
+
+[[tool.bumpversion.files]]
+filename = "Cargo.toml"
+search = "\nversion = \"{current_version}\""
+replace = "\nversion = \"{new_version}\""

python/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "lancedb-python"
-version = "0.4.10"
+version = "0.8.0"
 edition.workspace = true
 description = "Python bindings for LanceDB"
 license.workspace = true

python/pyproject.toml

@@ -1,9 +1,9 @@
 [project]
 name = "lancedb"
-version = "0.6.13"
+# version in Cargo.toml
 dependencies = [
     "deprecation",
-    "pylance==0.10.12",
+    "pylance==0.11.0",
     "ratelimiter~=1.0",
     "requests>=2.31.0",
     "retry>=0.9.2",

python/python/lancedb/_lancedb.pyi

@@ -86,3 +86,17 @@ class VectorQuery:
     def refine_factor(self, refine_factor: int): ...
     def nprobes(self, nprobes: int): ...
     def bypass_vector_index(self): ...
+
+class CompactionStats:
+    fragments_removed: int
+    fragments_added: int
+    files_removed: int
+    files_added: int
+
+class RemovalStats:
+    bytes_removed: int
+    old_versions_removed: int
+
+class OptimizeStats:
+    compaction: CompactionStats
+    prune: RemovalStats

python/python/lancedb/table.py

@@ -58,7 +58,7 @@ if TYPE_CHECKING:
     import PIL
     from lance.dataset import CleanupStats, ReaderLike
 
-    from ._lancedb import Table as LanceDBTable
+    from ._lancedb import Table as LanceDBTable, OptimizeStats
     from .db import LanceDBConnection
     from .index import BTree, IndexConfig, IvfPq
 
@@ -2377,6 +2377,49 @@ class AsyncTable:
         """
         await self._inner.restore()
 
+    async def optimize(
+        self, *, cleanup_older_than: Optional[timedelta] = None
+    ) -> OptimizeStats:
+        """
+        Optimize the on-disk data and indices for better performance.
+
+        Modeled after ``VACUUM`` in PostgreSQL.
+
+        Optimization covers three operations:
+
+         * Compaction: Merges small files into larger ones
+         * Prune: Removes old versions of the dataset
+         * Index: Optimizes the indices, adding new data to existing indices
+
+        Parameters
+        ----------
+        cleanup_older_than: timedelta, optional default 7 days
+            All files belonging to versions older than this will be removed.  Set
+            to 0 days to remove all versions except the latest.  The latest version
+            is never removed.
+
+        Experimental API
+        ----------------
+
+        The optimization process is undergoing active development and may change.
+        Our goal with these changes is to improve the performance of optimization and
+        reduce the complexity.
+
+        That being said, it is essential today to run optimize if you want the best
+        performance.  It should be stable and safe to use in production, but it our
+        hope that the API may be simplified (or not even need to be called) in the
+        future.
+
+        The frequency an application shoudl call optimize is based on the frequency of
+        data modifications.  If data is frequently added, deleted, or updated then
+        optimize should be run frequently.  A good rule of thumb is to run optimize if
+        you have added or modified 100,000 or more records or run more than 20 data
+        modification operations.
+        """
+        if cleanup_older_than is not None:
+            cleanup_older_than = round(cleanup_older_than.total_seconds() * 1000)
+        return await self._inner.optimize(cleanup_older_than)
+
     async def list_indices(self) -> IndexConfig:
         """
         List all indices that have been created with Self::create_index

python/python/tests/test_table.py

@@ -1025,3 +1025,29 @@ async def test_time_travel(db_async: AsyncConnection):
     # Can't use restore if not checked out
     with pytest.raises(ValueError, match="checkout before running restore"):
         await table.restore()
+
+
+@pytest.mark.asyncio
+async def test_optimize(db_async: AsyncConnection):
+    table = await db_async.create_table(
+        "test",
+        data=[{"x": [1]}],
+    )
+    await table.add(
+        data=[
+            {"x": [2]},
+        ],
+    )
+    stats = await table.optimize()
+    assert stats.compaction.files_removed == 2
+    assert stats.compaction.files_added == 1
+    assert stats.compaction.fragments_added == 1
+    assert stats.compaction.fragments_removed == 2
+    assert stats.prune.bytes_removed == 0
+    assert stats.prune.old_versions_removed == 0
+
+    stats = await table.optimize(cleanup_older_than=timedelta(seconds=0))
+    assert stats.prune.bytes_removed > 0
+    assert stats.prune.old_versions_removed == 3
+
+    assert await table.query().to_arrow() == pa.table({"x": [[1], [2]]})

python/src/table.rs

@@ -2,7 +2,9 @@ use arrow::{
     ffi_stream::ArrowArrayStreamReader,
     pyarrow::{FromPyArrow, ToPyArrow},
 };
-use lancedb::table::{AddDataMode, Table as LanceDbTable};
+use lancedb::table::{
+    AddDataMode, Duration, OptimizeAction, OptimizeOptions, Table as LanceDbTable,
+};
 use pyo3::{
     exceptions::{PyRuntimeError, PyValueError},
     pyclass, pymethods,
@@ -17,6 +19,40 @@ use crate::{
     query::Query,
 };
 
+/// Statistics about a compaction operation.
+#[pyclass(get_all)]
+#[derive(Clone, Debug)]
+pub struct CompactionStats {
+    /// The number of fragments removed
+    pub fragments_removed: u64,
+    /// The number of new, compacted fragments added
+    pub fragments_added: u64,
+    /// The number of data files removed
+    pub files_removed: u64,
+    /// The number of new, compacted data files added
+    pub files_added: u64,
+}
+
+/// Statistics about a cleanup operation
+#[pyclass(get_all)]
+#[derive(Clone, Debug)]
+pub struct RemovalStats {
+    /// The number of bytes removed
+    pub bytes_removed: u64,
+    /// The number of old versions removed
+    pub old_versions_removed: u64,
+}
+
+/// Statistics about an optimize operation
+#[pyclass(get_all)]
+#[derive(Clone, Debug)]
+pub struct OptimizeStats {
+    /// Statistics about the compaction operation
+    pub compaction: CompactionStats,
+    /// Statistics about the removal operation
+    pub prune: RemovalStats,
+}
+
 #[pyclass]
 pub struct Table {
     // We keep a copy of the name to use if the inner table is dropped
@@ -191,4 +227,58 @@ impl Table {
     pub fn query(&self) -> Query {
         Query::new(self.inner_ref().unwrap().query())
     }
+
+    pub fn optimize(self_: PyRef<'_, Self>, cleanup_since_ms: Option<u64>) -> PyResult<&PyAny> {
+        let inner = self_.inner_ref()?.clone();
+        let older_than = if let Some(ms) = cleanup_since_ms {
+            if ms > i64::MAX as u64 {
+                return Err(PyValueError::new_err(format!(
+                    "cleanup_since_ms must be between {} and -{}",
+                    i32::MAX,
+                    i32::MAX
+                )));
+            }
+            Duration::try_milliseconds(ms as i64)
+        } else {
+            None
+        };
+        future_into_py(self_.py(), async move {
+            let compaction_stats = inner
+                .optimize(OptimizeAction::Compact {
+                    options: lancedb::table::CompactionOptions::default(),
+                    remap_options: None,
+                })
+                .await
+                .infer_error()?
+                .compaction
+                .unwrap();
+            let prune_stats = inner
+                .optimize(OptimizeAction::Prune {
+                    older_than,
+                    delete_unverified: None,
+                })
+                .await
+                .infer_error()?
+                .prune
+                .unwrap();
+            inner
+                .optimize(lancedb::table::OptimizeAction::Index(
+                    OptimizeOptions::default(),
+                ))
+                .await
+                .infer_error()?;
+            Ok(OptimizeStats {
+                compaction: CompactionStats {
+                    files_added: compaction_stats.files_added as u64,
+                    files_removed: compaction_stats.files_removed as u64,
+                    fragments_added: compaction_stats.fragments_added as u64,
+                    fragments_removed: compaction_stats.fragments_removed as u64,
+                },
+                prune: RemovalStats {
+                    bytes_removed: prune_stats.bytes_removed,
+                    old_versions_removed: prune_stats.old_versions,
+                },
+            })
+        })
+    }
 }

release_process.md

@@ -8,6 +8,51 @@ The Python package is versioned and released separately from the Rust and Node.j
 ones. For Rust and Node.js, the release process is shared between `lancedb` and
 `vectordb` for now.
 
+## Preview releases
+
+LanceDB has full releases about every 2 weeks, but in between we make frequent
+preview releases. These are released as `0.x.y.betaN` versions. They receive the
+same level of testing as normal releases and let you get access to the latest
+features. However, we do not guarantee that preview releases will be available
+more than 6 months after they are released. We may delete the preview releases
+from the packaging index after a while. Once your application is stable, we
+recommend switching to full releases, which will never be removed from package
+indexes.
+
+## Making releases
+
+The release process uses a handful of GitHub actions to automate the process.
+
+```text
+  ┌─────────────────────┐                                                
+  │Create Release Commit│                                                
+  └─┬───────────────────┘                                                
+    │                           ┌────────────┐ ┌──►Python GH Release     
+    ├──►(tag) python-vX.Y.Z ───►│PyPI Publish├─┤                         
+    │                           └────────────┘ └──►Python Wheels         
+    │                                                                    
+    │                           ┌───────────┐                            
+    └──►(tag) vX.Y.Z ───┬──────►│NPM Publish├──┬──►Rust/Node GH Release  
+                        │       └───────────┘  │                         
+                        │                      └──►NPM Packages          
+                        │       ┌─────────────┐                          
+                        └──────►│Cargo Publish├───►Cargo Release         
+                                └─────────────┘                          
+```
+
+To start a release, trigger a `Create Release Commit` action from
+[the workflows page](https://github.com/lancedb/lancedb/actions/workflows/make-release-commit.yml)
+(Click on "Run workflow").
+
+* **For a preview release**, leave the default parameters.
+* **For a stable release**, set the `release_type` input to `stable`.
+
+> [!IMPORTANT]
+> If there was a breaking change since the last stable release, and we haven't
+> done so yet, we should increment the minor version. The CI will detect if this
+> is needed and fail the `Create Release Commit` job. To fix, select the
+> "bump minor version" option.
+
 ## Breaking changes
 
 We try to avoid breaking changes, but sometimes they are necessary. When there
@@ -21,12 +66,10 @@ body of the PR. A CI job will add a `breaking-change` label to the PR, which is
 what will ultimately be used to CI to determine if the minor version should be
 incremented.
 
-A CI job will validate that if a `breaking-change` label is added, the minor
+> [!IMPORTANT]
-version is incremented in the `Cargo.toml` and `pyproject.toml` files. The only
+> Reviewers should check that PRs with breaking changes receive the `breaking-change`
-exception is if it has already been incremented since the last stable release.
+> label. If a PR is missing the label, please add it, even if after it was merged.
-
+> This label is used in the release process.
-**It is the responsibility of the PR author to increment the minor version when
-appropriate.**
 
 Some things that are considered breaking changes:
 

rust/ffi/node/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "lancedb-node"
-version = "0.4.20"
+version = "0.5.0"
 description = "Serverless, low-latency vector database for AI applications"
 license.workspace = true
 edition.workspace = true

rust/ffi/node/src/error.rs

@@ -19,10 +19,12 @@ use snafu::Snafu;
 
 #[derive(Debug, Snafu)]
 pub enum Error {
+    #[allow(dead_code)]
     #[snafu(display("column '{name}' is missing"))]
     MissingColumn { name: String },
     #[snafu(display("{name}: {message}"))]
     OutOfRange { name: String, message: String },
+    #[allow(dead_code)]
     #[snafu(display("{index_type} is not a valid index type"))]
     InvalidIndexType { index_type: String },
 

rust/ffi/node/src/neon_ext/js_object_ext.rs

@@ -19,6 +19,7 @@ use neon::prelude::*;
 pub trait JsObjectExt {
     fn get_opt_u32(&self, cx: &mut FunctionContext, key: &str) -> Result<Option<u32>>;
     fn get_usize(&self, cx: &mut FunctionContext, key: &str) -> Result<usize>;
+    #[allow(dead_code)]
     fn get_opt_usize(&self, cx: &mut FunctionContext, key: &str) -> Result<Option<usize>>;
 }
 

rust/ffi/node/src/table.rs

@@ -324,7 +324,7 @@ impl JsTable {
         rt.spawn(async move {
             let stats = table
                 .optimize(OptimizeAction::Prune {
-                    older_than,
+                    older_than: Some(older_than),
                     delete_unverified,
                 })
                 .await;

rust/lancedb/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "lancedb"
-version = "0.4.20"
+version = "0.5.0"
 edition.workspace = true
 description = "LanceDB: A serverless, low-latency vector database for AI applications"
 license.workspace = true
@@ -40,8 +40,8 @@ serde = { version = "^1" }
 serde_json = { version = "1" }
 # For remote feature
 reqwest = { version = "0.11.24", features = ["gzip", "json"], optional = true }
-polars-arrow = { version = ">=0.37", optional = true }
+polars-arrow = { version = ">=0.37,<0.40.0", optional = true }
-polars = { version = ">=0.37", optional = true}
+polars = { version = ">=0.37,<0.40.0", optional = true}
 
 [dev-dependencies]
 tempfile = "3.5.0"
@@ -49,9 +49,12 @@ rand = { version = "0.8.3", features = ["small_rng"] }
 uuid = { version = "1.7.0", features = ["v4"] }
 walkdir = "2"
 # For s3 integration tests (dev deps aren't allowed to be optional atm)
-aws-sdk-s3 = { version = "1.0" }
+# We pin these because the content-length check breaks with localstack
-aws-sdk-kms = { version = "1.0" }
+# https://github.com/smithy-lang/smithy-rs/releases/tag/release-2024-05-21
+aws-sdk-s3 = { version = "=1.23.0" }
+aws-sdk-kms = { version = "=1.21.0" }
 aws-config = { version = "1.0" }
+aws-smithy-runtime = { version = "=1.3.0" }
 
 [features]
 default = []

rust/lancedb/src/connection.rs

@@ -195,7 +195,7 @@ impl<T: IntoArrow> CreateTableBuilder<true, T> {
             .embedding_registry()
             .get(&definition.embedding_name)
             .ok_or_else(|| Error::EmbeddingFunctionNotFound {
-                name: definition.embedding_name.to_string(),
+                name: definition.embedding_name.clone(),
                 reason: "No embedding function found in the connection's embedding_registry"
                     .to_string(),
             })?;

rust/lancedb/src/embeddings.rs

@@ -155,7 +155,7 @@ impl<R: RecordBatchReader> MaybeEmbedded<R> {
                         }
                         None => {
                             return Err(Error::EmbeddingFunctionNotFound {
-                                name: embedding_def.embedding_name.to_string(),
+                                name: embedding_def.embedding_name.clone(),
                                 reason: format!(
                                     "Table was defined with an embedding column `{}` but no embedding function was found with that name within the registry.",
                                     embedding_def.embedding_name

rust/lancedb/src/index.rs

@@ -16,7 +16,10 @@ use std::sync::Arc;
 
 use crate::{table::TableInternal, Result};
 
-use self::{scalar::BTreeIndexBuilder, vector::IvfPqIndexBuilder};
+use self::{
+    scalar::BTreeIndexBuilder,
+    vector::{IvfHnswSqIndexBuilder, IvfPqIndexBuilder},
+};
 
 pub mod scalar;
 pub mod vector;
@@ -25,6 +28,7 @@ pub enum Index {
     Auto,
     BTree(BTreeIndexBuilder),
     IvfPq(IvfPqIndexBuilder),
+    IvfHnswSq(IvfHnswSqIndexBuilder),
 }
 
 /// Builder for the create_index operation
@@ -65,6 +69,7 @@ impl IndexBuilder {
 #[derive(Debug, Clone, PartialEq)]
 pub enum IndexType {
     IvfPq,
+    IvfHnswSq,
     BTree,
 }
 

rust/lancedb/src/index/vector.rs

@@ -83,10 +83,14 @@ pub struct VectorIndexStatistics {
 #[derive(Debug, Clone)]
 pub struct IvfPqIndexBuilder {
     pub(crate) distance_type: DistanceType,
+
+    // IVF
     pub(crate) num_partitions: Option<u32>,
-    pub(crate) num_sub_vectors: Option<u32>,
     pub(crate) sample_rate: u32,
     pub(crate) max_iterations: u32,
+
+    // PQ
+    pub(crate) num_sub_vectors: Option<u32>,
 }
 
 impl Default for IvfPqIndexBuilder {
@@ -201,3 +205,124 @@ pub(crate) fn suggested_num_sub_vectors(dim: u32) -> u32 {
         1
     }
 }
+
+/// Builder for an IVF_HNSW_SQ index.
+///
+/// This index is a combination of IVF and HNSW.
+/// The IVF part is the same as the IVF PQ index.
+/// For each IVF partition, this builds a HNSW graph, the graph is used to
+/// quickly find the closest vectors to a query vector.
+///
+/// The SQ (scalar quantizer) is used to compress the vectors,
+/// each vector is mapped to a 8-bit integer vector, 4x compression ratio for float32 vector.
+#[derive(Debug, Clone)]
+pub struct IvfHnswSqIndexBuilder {
+    // IVF
+    pub(crate) distance_type: DistanceType,
+    pub(crate) num_partitions: Option<u32>,
+    pub(crate) sample_rate: u32,
+    pub(crate) max_iterations: u32,
+
+    // HNSW
+    pub(crate) m: u32,
+    pub(crate) ef_construction: u32,
+    // SQ
+    // TODO add num_bits for SQ after it supports another num_bits besides 8
+}
+
+impl Default for IvfHnswSqIndexBuilder {
+    fn default() -> Self {
+        Self {
+            distance_type: DistanceType::L2,
+            num_partitions: None,
+            sample_rate: 256,
+            max_iterations: 50,
+            m: 20,
+            ef_construction: 300,
+        }
+    }
+}
+
+impl IvfHnswSqIndexBuilder {
+    /// [DistanceType] to use to build the index.
+    ///
+    /// Default value is [DistanceType::L2].
+    ///
+    /// This is used when training the index to calculate the IVF partitions (vectors are
+    /// grouped in partitions with similar vectors according to this distance type)
+    ///
+    /// The metric type used to train an index MUST match the metric type used to search the
+    /// index.  Failure to do so will yield inaccurate results.
+    ///
+    /// Now IVF_HNSW_SQ only supports L2 and Cosine distance types.
+    pub fn distance_type(mut self, distance_type: DistanceType) -> Self {
+        self.distance_type = distance_type;
+        self
+    }
+
+    /// The number of IVF partitions to create.
+    ///
+    /// This value should generally scale with the number of rows in the dataset.  By default
+    /// the number of partitions is the square root of the number of rows.
+    ///
+    /// If this value is too large then the first part of the search (picking the right partition)
+    /// will be slow.  If this value is too small then the second part of the search (searching
+    /// within a partition) will be slow.
+    pub fn num_partitions(mut self, num_partitions: u32) -> Self {
+        self.num_partitions = Some(num_partitions);
+        self
+    }
+
+    /// The rate used to calculate the number of training vectors for kmeans and SQ.
+    ///
+    /// When an IVF_HNSW_SQ index is trained, we need to calculate partitions and min/max value of vectors.  These are groups
+    /// of vectors that are similar to each other.  To do this we use an algorithm called kmeans.
+    ///
+    /// Running kmeans on a large dataset can be slow.  To speed this up we run kmeans on a
+    /// random sample of the data.  This parameter controls the size of the sample.  The total
+    /// number of vectors used to train the IVF is `sample_rate * num_partitions`.
+    ///
+    /// The total number of vectors used to train the SQ is `sample_rate * 2^{num_bits}`.
+    ///
+    /// Increasing this value might improve the quality of the index but in most cases the
+    /// default should be sufficient.
+    ///
+    /// The default value is 256.
+    pub fn sample_rate(mut self, sample_rate: u32) -> Self {
+        self.sample_rate = sample_rate;
+        self
+    }
+
+    /// Max iterations to train kmeans.
+    ///
+    /// When training an IVF index we use kmeans to calculate the partitions.  This parameter
+    /// controls how many iterations of kmeans to run.
+    ///
+    /// Increasing this might improve the quality of the index but in most cases the parameter
+    /// is unused because kmeans will converge with fewer iterations.  The parameter is only
+    /// used in cases where kmeans does not appear to converge.  In those cases it is unlikely
+    /// that setting this larger will lead to the index converging anyways.
+    ///
+    /// The default value is 50.
+    pub fn max_iterations(mut self, max_iterations: u32) -> Self {
+        self.max_iterations = max_iterations;
+        self
+    }
+
+    /// The number of neighbors to select for each vector in the HNSW graph.
+    /// Bumping this number will increase the recall of the search but also increase the build/search time.
+    /// The default value is 20.
+    pub fn m(mut self, m: u32) -> Self {
+        self.m = m;
+        self
+    }
+
+    /// The number of candidates to evaluate during the construction of the HNSW graph.
+    /// Bumping this number will increase the recall of the search but also increase the build/search time.
+    /// This value should be not less than `ef` in the search phase.
+    /// The default value is 300.
+    pub fn ef_construction(mut self, ef_construction: u32) -> Self {
+        self.ef_construction = ef_construction;
+        self
+    }
+}

rust/lancedb/src/table.rs

@@ -23,12 +23,9 @@ use arrow::datatypes::Float32Type;
 use arrow_array::{RecordBatchIterator, RecordBatchReader};
 use arrow_schema::{DataType, Field, Schema, SchemaRef};
 use async_trait::async_trait;
-use chrono::Duration;
 use lance::dataset::builder::DatasetBuilder;
 use lance::dataset::cleanup::RemovalStats;
-use lance::dataset::optimize::{
+use lance::dataset::optimize::{compact_files, CompactionMetrics, IndexRemapperOptions};
-    compact_files, CompactionMetrics, CompactionOptions, IndexRemapperOptions,
-};
 use lance::dataset::scanner::{DatasetRecordBatchStream, Scanner};
 pub use lance::dataset::ColumnAlteration;
 pub use lance::dataset::NewColumnTransform;
@@ -38,8 +35,11 @@ use lance::dataset::{
 };
 use lance::dataset::{MergeInsertBuilder as LanceMergeInsertBuilder, WhenNotMatchedBySource};
 use lance::io::WrappingObjectStore;
+use lance_index::vector::hnsw::builder::HnswBuildParams;
+use lance_index::vector::ivf::IvfBuildParams;
+use lance_index::vector::sq::builder::SQBuildParams;
+use lance_index::DatasetIndexExt;
 use lance_index::IndexType;
-use lance_index::{optimize::OptimizeOptions, DatasetIndexExt};
 use log::info;
 use serde::{Deserialize, Serialize};
 use snafu::whatever;
@@ -48,7 +48,9 @@ use crate::arrow::IntoArrow;
 use crate::connection::NoData;
 use crate::embeddings::{EmbeddingDefinition, EmbeddingRegistry, MaybeEmbedded, MemoryRegistry};
 use crate::error::{Error, Result};
-use crate::index::vector::{IvfPqIndexBuilder, VectorIndex, VectorIndexStatistics};
+use crate::index::vector::{
+    IvfHnswSqIndexBuilder, IvfPqIndexBuilder, VectorIndex, VectorIndexStatistics,
+};
 use crate::index::IndexConfig;
 use crate::index::{
     vector::{suggested_num_partitions, suggested_num_sub_vectors},
@@ -65,6 +67,10 @@ use self::merge::MergeInsertBuilder;
 pub(crate) mod dataset;
 pub mod merge;
 
+pub use chrono::Duration;
+pub use lance::dataset::optimize::CompactionOptions;
+pub use lance_index::optimize::OptimizeOptions;
+
 /// Defines the type of column
 #[derive(Debug, Clone, Serialize, Deserialize)]
 pub enum ColumnKind {
@@ -145,22 +151,58 @@ impl TableDefinition {
 ///
 /// By default, it optimizes everything, as [`OptimizeAction::All`].
 pub enum OptimizeAction {
-    /// Run optimization on every, with default options.
+    /// Run all optimizations with default values
     All,
-    /// Compact files in the dataset
+    /// Compacts files in the dataset
+    ///
+    /// LanceDb uses a readonly filesystem for performance and safe concurrency.  Every time
+    /// new data is added it will be added into new files.  Small files
+    /// can hurt both read and write performance.  Compaction will merge small files
+    /// into larger ones.
+    ///
+    /// All operations that modify data (add, delete, update, merge insert, etc.) will create
+    /// new files.  If these operations are run frequently then compaction should run frequently.
+    ///
+    /// If these operations are never run (search only) then compaction is not necessary.
     Compact {
         options: CompactionOptions,
         remap_options: Option<Arc<dyn IndexRemapperOptions>>,
     },
-    /// Prune old version of datasets.
+    /// Prune old version of datasets
+    ///
+    /// Every change in LanceDb is additive.  When data is removed from a dataset a new version is
+    /// created that doesn't contain the removed data.  However, the old version, which does contain
+    /// the removed data, is left in place.  This is necessary for consistency and concurrency and
+    /// also enables time travel functionality like the ability to checkout an older version of the
+    /// dataset to undo changes.
+    ///
+    /// Over time, these old versions can consume a lot of disk space.  The prune operation will
+    /// remove versions of the dataset that are older than a certain age.  This will free up the
+    /// space used by that old data.
+    ///
+    /// Once a version is pruned it can no longer be checked out.
     Prune {
         /// The duration of time to keep versions of the dataset.
-        older_than: Duration,
+        older_than: Option<Duration>,
         /// Because they may be part of an in-progress transaction, files newer than 7 days old are not deleted by default.
         /// If you are sure that there are no in-progress transactions, then you can set this to True to delete all files older than `older_than`.
         delete_unverified: Option<bool>,
     },
-    /// Optimize index.
+    /// Optimize the indices
+    ///
+    /// This operation optimizes all indices in the table.  When new data is added to LanceDb
+    /// it is not added to the indices.  However, it can still turn up in searches because the search
+    /// function will scan both the indexed data and the unindexed data in parallel.  Over time, the
+    /// unindexed data can become large enough that the search performance is slow.  This operation
+    /// will add the unindexed data to the indices without rerunning the full index creation process.
+    ///
+    /// Optimizing an index is faster than re-training the index but it does not typically adjust the
+    /// underlying model relied upon by the index.  This can eventually lead to poor search accuracy
+    /// and so users may still want to occasionally retrain the index after adding a large amount of
+    /// data.
+    ///
+    /// For example, when using IVF, an index will create clusters.  Optimizing an index assigns unindexed
+    /// data to the existing clusters, but it does not move the clusters or create new clusters.
     Index(OptimizeOptions),
 }
 
@@ -312,6 +354,7 @@ impl UpdateBuilder {
 
 #[async_trait]
 pub(crate) trait TableInternal: std::fmt::Display + std::fmt::Debug + Send + Sync {
+    #[allow(dead_code)]
     fn as_any(&self) -> &dyn std::any::Any;
     /// Cast as [`NativeTable`], or return None it if is not a [`NativeTable`].
     fn as_native(&self) -> Option<&NativeTable>;
@@ -751,10 +794,30 @@ impl Table {
 
     /// Optimize the on-disk data and indices for better performance.
     ///
+    /// Modeled after ``VACUUM`` in PostgreSQL.
+    ///
+    /// Optimization is discussed in more detail in the [OptimizeAction] documentation
+    /// and covers three operations:
+    ///
+    ///  * Compaction: Merges small files into larger ones
+    ///  * Prune: Removes old versions of the dataset
+    ///  * Index: Optimizes the indices, adding new data to existing indices
+    ///
     /// <section class="warning">Experimental API</section>
     ///
-    /// Modeled after ``VACUUM`` in PostgreSQL.
+    /// The optimization process is undergoing active development and may change.
-    /// Not all implementations support explicit optimization.
+    /// Our goal with these changes is to improve the performance of optimization and
+    /// reduce the complexity.
+    ///
+    /// That being said, it is essential today to run optimize if you want the best
+    /// performance.  It should be stable and safe to use in production, but it our
+    /// hope that the API may be simplified (or not even need to be called) in the future.
+    ///
+    /// The frequency an application shoudl call optimize is based on the frequency of
+    /// data modifications.  If data is frequently added, deleted, or updated then
+    /// optimize should be run frequently.  A good rule of thumb is to run optimize if
+    /// you have added or modified 100,000 or more records or run more than 20 data
+    /// modification operations.
     pub async fn optimize(&self, action: OptimizeAction) -> Result<OptimizeStats> {
         self.inner.optimize(action).await
     }
@@ -1238,7 +1301,6 @@ impl NativeTable {
             num_partitions as usize,
             /*num_bits=*/ 8,
             num_sub_vectors as usize,
-            false,
             index.distance_type.into(),
             index.max_iterations as usize,
         );
@@ -1254,6 +1316,57 @@ impl NativeTable {
         Ok(())
     }
 
+    async fn create_ivf_hnsw_sq_index(
+        &self,
+        index: IvfHnswSqIndexBuilder,
+        field: &Field,
+        replace: bool,
+    ) -> Result<()> {
+        if !Self::supported_vector_data_type(field.data_type()) {
+            return Err(Error::InvalidInput {
+                message: format!(
+                    "An IVF HNSW SQ index cannot be created on the column `{}` which has data type {}",
+                    field.name(),
+                    field.data_type()
+                ),
+            });
+        }
+
+        let num_partitions = if let Some(n) = index.num_partitions {
+            n
+        } else {
+            suggested_num_partitions(self.count_rows(None).await?)
+        };
+
+        let mut dataset = self.dataset.get_mut().await?;
+        let mut ivf_params = IvfBuildParams::new(num_partitions as usize);
+        ivf_params.sample_rate = index.sample_rate as usize;
+        ivf_params.max_iters = index.max_iterations as usize;
+        let hnsw_params = HnswBuildParams::default()
+            .num_edges(index.m as usize)
+            .ef_construction(index.ef_construction as usize);
+        let sq_params = SQBuildParams {
+            sample_rate: index.sample_rate as usize,
+            ..Default::default()
+        };
+        let lance_idx_params = lance::index::vector::VectorIndexParams::with_ivf_hnsw_sq_params(
+            index.distance_type.into(),
+            ivf_params,
+            hnsw_params,
+            sq_params,
+        );
+        dataset
+            .create_index(
+                &[field.name()],
+                IndexType::Vector,
+                None,
+                &lance_idx_params,
+                replace,
+            )
+            .await?;
+        Ok(())
+    }
+
     async fn create_auto_index(&self, field: &Field, opts: IndexBuilder) -> Result<()> {
         if Self::supported_vector_data_type(field.data_type()) {
             self.create_ivf_pq_index(IvfPqIndexBuilder::default(), field, opts.replace)
@@ -1497,6 +1610,10 @@ impl TableInternal for NativeTable {
             Index::Auto => self.create_auto_index(field, opts).await,
             Index::BTree(_) => self.create_btree_index(field, opts).await,
             Index::IvfPq(ivf_pq) => self.create_ivf_pq_index(ivf_pq, field, opts.replace).await,
+            Index::IvfHnswSq(ivf_hnsw_sq) => {
+                self.create_ivf_hnsw_sq_index(ivf_hnsw_sq, field, opts.replace)
+                    .await
+            }
         }
     }
 
@@ -1592,7 +1709,7 @@ impl TableInternal for NativeTable {
                     .compaction;
                 stats.prune = self
                     .optimize(OptimizeAction::Prune {
-                        older_than: Duration::try_days(7).unwrap(),
+                        older_than: None,
                         delete_unverified: None,
                     })
                     .await?
@@ -1611,8 +1728,11 @@ impl TableInternal for NativeTable {
                 delete_unverified,
             } => {
                 stats.prune = Some(
-                    self.cleanup_old_versions(older_than, delete_unverified)
+                    self.cleanup_old_versions(
-                        .await?,
+                        older_than.unwrap_or(Duration::try_days(7).expect("valid delta")),
+                        delete_unverified,
+                    )
+                    .await?,
                 );
             }
             OptimizeAction::Index(options) => {
@@ -2357,6 +2477,102 @@ mod tests {
         );
     }
 
+    #[tokio::test]
+    async fn test_create_index_ivf_hnsw_sq() {
+        use arrow_array::RecordBatch;
+        use arrow_schema::{DataType, Field, Schema as ArrowSchema};
+        use rand;
+        use std::iter::repeat_with;
+
+        use arrow_array::Float32Array;
+
+        let tmp_dir = tempdir().unwrap();
+        let uri = tmp_dir.path().to_str().unwrap();
+        let conn = connect(uri).execute().await.unwrap();
+
+        let dimension = 16;
+        let schema = Arc::new(ArrowSchema::new(vec![Field::new(
+            "embeddings",
+            DataType::FixedSizeList(
+                Arc::new(Field::new("item", DataType::Float32, true)),
+                dimension,
+            ),
+            false,
+        )]));
+
+        let mut rng = rand::thread_rng();
+        let float_arr = Float32Array::from(
+            repeat_with(|| rng.gen::<f32>())
+                .take(512 * dimension as usize)
+                .collect::<Vec<f32>>(),
+        );
+
+        let vectors = Arc::new(create_fixed_size_list(float_arr, dimension).unwrap());
+        let batches = RecordBatchIterator::new(
+            vec![RecordBatch::try_new(schema.clone(), vec![vectors.clone()]).unwrap()]
+                .into_iter()
+                .map(Ok),
+            schema,
+        );
+
+        let table = conn.create_table("test", batches).execute().await.unwrap();
+
+        assert_eq!(
+            table
+                .as_native()
+                .unwrap()
+                .count_indexed_rows("my_index")
+                .await
+                .unwrap(),
+            None
+        );
+        assert_eq!(
+            table
+                .as_native()
+                .unwrap()
+                .count_unindexed_rows("my_index")
+                .await
+                .unwrap(),
+            None
+        );
+
+        let index = IvfHnswSqIndexBuilder::default();
+        table
+            .create_index(&["embeddings"], Index::IvfHnswSq(index))
+            .execute()
+            .await
+            .unwrap();
+
+        let index_configs = table.list_indices().await.unwrap();
+        assert_eq!(index_configs.len(), 1);
+        let index = index_configs.into_iter().next().unwrap();
+        assert_eq!(index.index_type, crate::index::IndexType::IvfPq);
+        assert_eq!(index.columns, vec!["embeddings".to_string()]);
+        assert_eq!(table.count_rows(None).await.unwrap(), 512);
+        assert_eq!(table.name(), "test");
+
+        let indices = table.as_native().unwrap().load_indices().await.unwrap();
+        let index_uuid = &indices[0].index_uuid;
+        assert_eq!(
+            table
+                .as_native()
+                .unwrap()
+                .count_indexed_rows(index_uuid)
+                .await
+                .unwrap(),
+            Some(512)
+        );
+        assert_eq!(
+            table
+                .as_native()
+                .unwrap()
+                .count_unindexed_rows(index_uuid)
+                .await
+                .unwrap(),
+            Some(0)
+        );
+    }
+
     fn create_fixed_size_list<T: Array>(values: T, list_size: i32) -> Result<FixedSizeListArray> {
         let list_type = DataType::FixedSizeList(
             Arc::new(Field::new("item", values.data_type().clone(), true)),