refactor: forward deprecated replace_field_metadata to update_field_metadata

### Summary
Adds update_field_metadata to the client SDK (Rust core, Python, and
TypeScript) so clients can edit per-field (column) Arrow metadata
(schema.fields[].metadata)

### Testing
- added unit tests
- ran E2E against a local server on both local and remote tables (set →
merge → delete), across Python sync/async and TypeScript

### Next steps
- deprecate replace_field_metadata in the python lancedb favor of this
(typescript didn't have replace_field_metadata method). This matches
Lance's API direction (Lance already deprecated replace_field_metadata
for update_field_metadata)

.agents/skills/README.md

@@ -0,0 +1,7 @@
+# Agent Skills
+
+This directory contains repo-scoped code agent skills for the LanceDB project.
+
+Each skill is a folder that contains a required `SKILL.md` and optional bundled resources.
+
+Codex discovers skills from `.agents/skills` in the current working directory and parent directories.

.agents/skills/lancedb-update-lance-dependency/SKILL.md

@@ -0,0 +1,98 @@
+---
+name: lancedb-update-lance-dependency
+description: Update LanceDB to a specific Lance release or tag. Use when bumping Lance dependencies in the lancedb repository, including Rust workspace Lance crates, Java lance-core, validation, branch creation, commit, push, and PR creation when requested.
+---
+
+# LanceDB Update Lance Dependency
+
+## Scope
+
+Use this skill in the `lancedb/lancedb` repository when updating the Lance dependency to a specific Lance version or tag.
+
+Inputs can be a version (`7.2.0-beta.1`), a tag (`v7.2.0-beta.1`), a tag ref (`refs/tags/v7.2.0-beta.1`), or `latest`.
+
+## Workflow
+
+1. Confirm the worktree status with `git status --short`.
+2. Resolve the target Lance version:
+
+   - If the input is `latest`, empty, or omitted, run:
+
+     ```bash
+     python3 ci/check_lance_release.py
+     ```
+
+     Parse the JSON output. If `needs_update` is not `true`, stop without creating a PR. Otherwise use `latest_tag`.
+
+   - If the input is explicit, use it directly.
+
+3. Compute update metadata without changing files:
+
+   ```bash
+   python3 ci/update_lance_dependency.py "$TAG_OR_VERSION" --metadata-only
+   ```
+
+   Before making changes, check for an existing open PR with the emitted `pr_title`:
+
+   ```bash
+   gh pr list --search "\"$PR_TITLE\" in:title" --state open --limit 1 --json number,url,title
+   ```
+
+   If a matching open PR exists, stop and report it instead of creating a duplicate.
+
+4. Run the deterministic update entrypoint:
+
+   ```bash
+   python3 ci/update_lance_dependency.py "$TAG_OR_VERSION"
+   ```
+
+   This updates the Rust workspace Lance dependencies through `ci/set_lance_version.py`, updates `java/pom.xml`, refreshes Cargo metadata, and prints JSON metadata containing `branch_name`, `commit_message`, and `pr_title`.
+
+5. Run validation:
+
+   ```bash
+   cargo clippy --quiet --workspace --tests --all-features -- -D warnings
+   cargo fmt --all --quiet
+   ```
+
+   Fix real diagnostics and rerun clippy until it succeeds. Do not skip warnings.
+
+6. Inspect `git status --short` and `git diff` to ensure only the Lance dependency update and required compatibility fixes are present.
+
+7. If the task only asks to prepare local changes, stop here and report the changed files and validation result.
+
+8. If the task asks to publish the update, create a branch using the printed `branch_name`, stage all relevant files, and commit using the printed `commit_message`. Do not amend or rewrite existing commits.
+
+9. Push to `origin`. Before creating the PR, check that the current token has push permission:
+
+   ```bash
+   gh api repos/lancedb/lancedb --jq .permissions.push
+   ```
+
+   If the remote branch already exists for the same generated branch name, delete the remote ref with `gh api -X DELETE repos/lancedb/lancedb/git/refs/heads/$BRANCH_NAME`, then push. Do not force-push.
+
+10. Create a PR targeting `main` with the printed `pr_title`. If there is no PR template, keep the body to two or three concise sentences: state the Lance dependency bump, note any required compatibility fixes, and link the triggering Lance tag or release.
+
+11. Read back the remote PR title after creation. If it is not a Conventional Commit title, fix it immediately.
+
+12. When running in GitHub Actions after creating the LanceDB PR, trigger the Sophon dependency update:
+
+    ```bash
+    gh workflow run codex-bump-lancedb-lance.yml \
+      --repo lancedb/sophon \
+      -f lance_ref="$LANCE_TAG" \
+      -f lancedb_ref="$BRANCH_NAME"
+    gh run list --repo lancedb/sophon --workflow codex-bump-lancedb-lance.yml --limit 1 --json databaseId,url,displayTitle
+    ```
+
+    Use the emitted metadata `tag` value as `LANCE_TAG`. Do this only after a new LanceDB PR has been created. If the update was skipped because no update is needed or an open PR already exists, do not trigger Sophon.
+
+## GitHub Actions
+
+When this skill is used from GitHub Actions, `TAG`, `GH_TOKEN`, and `GITHUB_TOKEN` may already be set. Resolve `latest` first when `TAG` is empty. Once an explicit tag or version is known, use:
+
+```bash
+python3 ci/update_lance_dependency.py "$TAG" --github-output "$GITHUB_OUTPUT"
+```
+
+Then use the emitted `branch_name`, `commit_message`, and `pr_title` values for branch, commit, and PR creation.

.bumpversion.toml

@@ -1,5 +1,5 @@
 [tool.bumpversion]
-current_version = "0.29.1-beta.0"
+current_version = "0.30.1-beta.0"
 parse = """(?x)
     (?P<major>0|[1-9]\\d*)\\.
     (?P<minor>0|[1-9]\\d*)\\.

.github/workflows/build_windows_wheel/action.yml

@@ -29,7 +29,3 @@ runs:
         args: ${{ inputs.args }}
         docker-options: "-e PIP_EXTRA_INDEX_URL='https://pypi.fury.io/lance-format/ https://pypi.fury.io/lancedb/'"
         working-directory: python
-    - uses: actions/upload-artifact@v4
-      with:
-        name: windows-wheels
-        path: python\target\wheels

.github/workflows/codex-update-lance-dependency.yml

@@ -4,14 +4,16 @@ on:
   workflow_call:
     inputs:
       tag:
-        description: "Tag name from Lance"
-        required: true
+        description: "Tag name from Lance. If omitted, the skill will use the latest Lance release that needs an update."
+        required: false
+        default: ""
         type: string
   workflow_dispatch:
     inputs:
       tag:
-        description: "Tag name from Lance"
-        required: true
+        description: "Tag name from Lance. Leave empty to use the latest Lance release that needs an update."
+        required: false
+        default: ""
         type: string
 
 permissions:
@@ -25,7 +27,7 @@ jobs:
     steps:
       - name: Show inputs
         run: |
-          echo "tag = ${{ inputs.tag }}"
+          echo "tag = ${{ inputs.tag || 'latest' }}"
 
       - name: Checkout Repo LanceDB
         uses: actions/checkout@v4
@@ -71,65 +73,21 @@ jobs:
           OPENAI_API_KEY: ${{ secrets.CODEX_TOKEN }}
         run: |
           set -euo pipefail
-          VERSION="${TAG#refs/tags/}"
-          VERSION="${VERSION#v}"
-          BRANCH_NAME="codex/update-lance-${VERSION//[^a-zA-Z0-9]/-}"
-
-          # Use "chore" for beta/rc versions, "feat" for stable releases
-          if [[ "${VERSION}" == *beta* ]] || [[ "${VERSION}" == *rc* ]]; then
-            COMMIT_TYPE="chore"
-          else
-            COMMIT_TYPE="feat"
-          fi
+          TARGET_TAG="${TAG:-latest}"
 
           cat <<EOF >/tmp/codex-prompt.txt
-          You are running inside the lancedb repository on a GitHub Actions runner. Update the Lance dependency to version ${VERSION} and prepare a pull request for maintainers to review.
+          You are running inside the lancedb repository on a GitHub Actions runner.
 
-          Follow these steps exactly:
-          1. Use script "ci/set_lance_version.py" to update Lance Rust dependencies. The script already refreshes Cargo metadata, so allow it to finish even if it takes time.
-          2. Update the Java lance-core dependency version in "java/pom.xml": change the "<lance-core.version>...</lance-core.version>" property to "${VERSION}".
-          3. Run "cargo clippy --workspace --tests --all-features -- -D warnings". If diagnostics appear, fix them yourself and rerun clippy until it exits cleanly. Do not skip any warnings.
-          4. After clippy succeeds, run "cargo fmt --all" to format the workspace.
-          5. Ensure the repository is clean except for intentional changes. Inspect "git status --short" and "git diff" to confirm the dependency update and any required fixes.
-          6. Create and switch to a new branch named "${BRANCH_NAME}" (replace any duplicated hyphens if necessary).
-          7. Stage all relevant files with "git add -A". Commit using the message "${COMMIT_TYPE}: update lance dependency to v${VERSION}".
-          8. Push the branch to origin. If the remote branch already exists, delete it first with "gh api -X DELETE repos/lancedb/lancedb/git/refs/heads/${BRANCH_NAME}" then push with "git push origin ${BRANCH_NAME}". Do NOT use "git push --force" or "git push -f".
-          9. env "GH_TOKEN" is available, use "gh" tools for github related operations like creating pull request.
-          10. Create a pull request targeting "main" with title "${COMMIT_TYPE}: update lance dependency to v${VERSION}". First, write the PR body to /tmp/pr-body.md using a heredoc (cat <<'EOF' > /tmp/pr-body.md). The body should summarize the dependency bump, clippy/fmt verification, and link the triggering tag (${TAG}). Then run "gh pr create --body-file /tmp/pr-body.md".
-          11. After creating the PR, display the PR URL, "git status --short", and a concise summary of the commands run and their results.
+          Use \$lancedb-update-lance-dependency with target "${TARGET_TAG}".
 
           Constraints:
-          - Use bash commands; avoid modifying GitHub workflow files other than through the scripted task above.
-          - Do not merge the PR.
-          - If any command fails, diagnose and fix the issue instead of aborting.
+          - Use env "GH_TOKEN" for GitHub operations.
+          - Do not merge the pull request.
+          - Do not force-push.
+          - Do not create a duplicate pull request if an open PR already exists for the target Lance version.
+          - If any command fails, diagnose and fix the root cause instead of aborting.
+          - After creating the PR, display the PR URL, "git status --short", and a concise summary of the commands run and their results.
           EOF
 
           printenv OPENAI_API_KEY | codex login --with-api-key
           codex --config shell_environment_policy.ignore_default_excludes=true exec --dangerously-bypass-approvals-and-sandbox "$(cat /tmp/codex-prompt.txt)"
-
-      - name: Trigger sophon dependency update
-        env:
-          TAG: ${{ inputs.tag }}
-          GH_TOKEN: ${{ secrets.ROBOT_TOKEN }}
-        run: |
-          set -euo pipefail
-          VERSION="${TAG#refs/tags/}"
-          VERSION="${VERSION#v}"
-          LANCEDB_BRANCH="codex/update-lance-${VERSION//[^a-zA-Z0-9]/-}"
-
-          echo "Triggering sophon workflow with:"
-          echo "  lance_ref: ${TAG#refs/tags/}"
-          echo "  lancedb_ref: ${LANCEDB_BRANCH}"
-
-          gh workflow run codex-bump-lancedb-lance.yml \
-            --repo lancedb/sophon \
-            -f lance_ref="${TAG#refs/tags/}" \
-            -f lancedb_ref="${LANCEDB_BRANCH}"
-
-      - name: Show latest sophon workflow run
-        env:
-          GH_TOKEN: ${{ secrets.ROBOT_TOKEN }}
-        run: |
-          set -euo pipefail
-          echo "Latest sophon workflow run:"
-          gh run list --repo lancedb/sophon --workflow codex-bump-lancedb-lance.yml --limit 1 --json databaseId,url,displayTitle

.github/workflows/lance-release-timer.yml

@@ -1,62 +0,0 @@
-name: Lance Release Timer
-
-on:
-  schedule:
-    - cron: "*/10 * * * *"
-  workflow_dispatch:
-
-permissions:
-  contents: read
-  actions: write
-
-concurrency:
-  group: lance-release-timer
-  cancel-in-progress: false
-
-jobs:
-  trigger-update:
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v4
-
-      - name: Check for new Lance tag
-        id: check
-        env:
-          GH_TOKEN: ${{ secrets.ROBOT_TOKEN }}
-        run: |
-          python3 ci/check_lance_release.py --github-output "$GITHUB_OUTPUT"
-
-      - name: Look for existing PR
-        if: steps.check.outputs.needs_update == 'true'
-        id: pr
-        env:
-          GH_TOKEN: ${{ secrets.ROBOT_TOKEN }}
-        run: |
-          set -euo pipefail
-          TITLE="chore: update lance dependency to v${{ steps.check.outputs.latest_version }}"
-          COUNT=$(gh pr list --search "\"$TITLE\" in:title" --state open --limit 1 --json number --jq 'length')
-          if [ "$COUNT" -gt 0 ]; then
-            echo "Open PR already exists for $TITLE"
-            echo "pr_exists=true" >> "$GITHUB_OUTPUT"
-          else
-            echo "No existing PR for $TITLE"
-            echo "pr_exists=false" >> "$GITHUB_OUTPUT"
-          fi
-
-      - name: Trigger codex update workflow
-        if: steps.check.outputs.needs_update == 'true' && steps.pr.outputs.pr_exists != 'true'
-        env:
-          GH_TOKEN: ${{ secrets.ROBOT_TOKEN }}
-        run: |
-          set -euo pipefail
-          TAG=${{ steps.check.outputs.latest_tag }}
-          gh workflow run codex-update-lance-dependency.yml -f tag=refs/tags/$TAG
-
-      - name: Show latest codex workflow run
-        if: steps.check.outputs.needs_update == 'true' && steps.pr.outputs.pr_exists != 'true'
-        env:
-          GH_TOKEN: ${{ secrets.ROBOT_TOKEN }}
-        run: |
-          set -euo pipefail
-          gh run list --workflow codex-update-lance-dependency.yml --limit 1 --json databaseId,url,displayTitle

.github/workflows/pypi-publish.yml

@@ -8,6 +8,9 @@ on:
     # This should trigger a dry run (we skip the final publish step)
     paths:
       - .github/workflows/pypi-publish.yml
+      - .github/workflows/build_linux_wheel/action.yml
+      - .github/workflows/build_mac_wheel/action.yml
+      - .github/workflows/build_windows_wheel/action.yml
       - Cargo.toml # Change in dependency frequently breaks builds
       - Cargo.lock
 
@@ -21,32 +24,21 @@ jobs:
   linux:
     name: Python ${{ matrix.config.platform }} manylinux${{ matrix.config.manylinux }}
     timeout-minutes: 60
-    permissions:
-      id-token: write
-      contents: read
     strategy:
       matrix:
         config:
-          - platform: x86_64
-            manylinux: "2_17"
-            extra_args: ""
-            runner: ubuntu-22.04
           - platform: x86_64
             manylinux: "2_28"
             extra_args: "--features fp16kernels"
             runner: ubuntu-22.04
-          - platform: aarch64
-            manylinux: "2_17"
-            extra_args: ""
-            # For successful fat LTO builds, we need a large runner to avoid OOM errors.
-            runner: ubuntu-2404-8x-arm64
+          # For successful fat LTO builds, we need a large runner to avoid OOM errors.
           - platform: aarch64
             manylinux: "2_28"
             extra_args: "--features fp16kernels"
             runner: ubuntu-2404-8x-arm64
     runs-on: ${{ matrix.config.runner }}
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
         with:
           fetch-depth: 0
           lfs: true
@@ -60,15 +52,14 @@ jobs:
           args: "--release --strip ${{ matrix.config.extra_args }}"
           arm-build: ${{ matrix.config.platform == 'aarch64' }}
           manylinux: ${{ matrix.config.manylinux }}
-      - uses: ./.github/workflows/upload_wheel
+      - uses: actions/upload-artifact@v7
         if: startsWith(github.ref, 'refs/tags/python-v')
         with:
-          fury_token: ${{ secrets.FURY_TOKEN }}
+          name: wheels-linux-${{ matrix.config.platform }}-${{ matrix.config.manylinux }}
+          path: target/wheels/lancedb-*.whl
+          if-no-files-found: error
   mac:
     timeout-minutes: 90
-    permissions:
-      id-token: write
-      contents: read
     runs-on: ${{ matrix.config.runner }}
     strategy:
       matrix:
@@ -78,7 +69,7 @@ jobs:
     env:
       MACOSX_DEPLOYMENT_TARGET: 10.15
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
         with:
           fetch-depth: 0
           lfs: true
@@ -90,18 +81,21 @@ jobs:
         with:
           python-minor-version: 10
           args: "--release --strip --target ${{ matrix.config.target }} --features fp16kernels"
-      - uses: ./.github/workflows/upload_wheel
+      - uses: actions/upload-artifact@v7
         if: startsWith(github.ref, 'refs/tags/python-v')
         with:
-          fury_token: ${{ secrets.FURY_TOKEN }}
+          name: wheels-mac-${{ matrix.config.target }}
+          path: target/wheels/lancedb-*.whl
+          if-no-files-found: error
   windows:
-    timeout-minutes: 60
-    permissions:
-      id-token: write
-      contents: read
+    timeout-minutes: 90
     runs-on: windows-latest
+    env:
+      # link.exe is single-threaded and the long pole on Windows builds. Use
+      # rustc's bundled lld-link instead.
+      CARGO_TARGET_X86_64_PC_WINDOWS_MSVC_LINKER: rust-lld
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
         with:
           fetch-depth: 0
           lfs: true
@@ -113,18 +107,70 @@ jobs:
         with:
           python-minor-version: 10
           args: "--release --strip"
-          vcpkg_token: ${{ secrets.VCPKG_GITHUB_PACKAGES }}
-      - uses: ./.github/workflows/upload_wheel
+      - uses: actions/upload-artifact@v7
         if: startsWith(github.ref, 'refs/tags/python-v')
         with:
-          fury_token: ${{ secrets.FURY_TOKEN }}
+          name: wheels-windows
+          path: target/wheels/lancedb-*.whl
+          if-no-files-found: error
+  publish:
+    name: Publish wheels
+    if: startsWith(github.ref, 'refs/tags/python-v')
+    needs: [linux, mac, windows]
+    runs-on: ubuntu-latest
+    permissions:
+      id-token: write
+      contents: read
+    steps:
+      - uses: actions/checkout@v6
+      - name: Download wheel artifacts
+        uses: actions/download-artifact@v8
+        with:
+          pattern: wheels-*
+          path: target/wheels
+          merge-multiple: true
+      - name: List wheels
+        run: ls -la target/wheels
+      - name: Choose repo
+        id: choose_repo
+        run: |
+          if [[ ${{ github.ref }} == *beta* ]]; then
+            echo "repo=fury" >> $GITHUB_OUTPUT
+          else
+            echo "repo=pypi" >> $GITHUB_OUTPUT
+          fi
+      - name: Publish to Fury
+        if: steps.choose_repo.outputs.repo == 'fury'
+        env:
+          FURY_TOKEN: ${{ secrets.FURY_TOKEN }}
+        run: |
+          shopt -s nullglob
+          WHEELS=(target/wheels/lancedb-*.whl)
+          if [[ ${#WHEELS[@]} -eq 0 ]]; then
+            echo "No wheels found in target/wheels/" >&2
+            exit 1
+          fi
+          for WHEEL in "${WHEELS[@]}"; do
+            echo "Uploading $WHEEL to Fury"
+            curl -f -F package=@"$WHEEL" "https://$FURY_TOKEN@push.fury.io/lancedb/"
+          done
+      # NOTE: pypa/gh-action-pypi-publish must be invoked directly from a
+      # workflow file, not from inside a composite action. When called from a
+      # composite, `github.action_repository` is empty (actions/runner#2473)
+      # and the action falls back to `github.repository`, producing a bogus
+      # `docker://ghcr.io/<repo>:<ref>` image reference that GHA tries to pull.
+      - name: Publish to PyPI
+        if: steps.choose_repo.outputs.repo == 'pypi'
+        uses: pypa/gh-action-pypi-publish@release/v1
+        with:
+          packages-dir: target/wheels/
   gh-release:
     if: startsWith(github.ref, 'refs/tags/python-v')
     runs-on: ubuntu-latest
     permissions:
       contents: write
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
         with:
           fetch-depth: 0
           lfs: true
@@ -187,13 +233,13 @@ jobs:
   report-failure:
     name: Report Workflow Failure
     runs-on: ubuntu-latest
-    needs: [linux, mac, windows]
+    needs: [linux, mac, windows, publish]
     permissions:
       contents: read
       issues: write
     if: always() && failure() && startsWith(github.ref, 'refs/tags/python-v')
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
       - uses: ./.github/actions/create-failure-issue
         with:
           job-results: ${{ toJSON(needs) }}

.github/workflows/upload_wheel/action.yml

@@ -1,34 +0,0 @@
-name: upload-wheel
-
-description: "Upload wheels to Pypi"
-inputs:
-  fury_token:
-    required: true
-    description: "release token for the fury repo"
-
-runs:
-  using: "composite"
-  steps:
-  - name: Choose repo
-    shell: bash
-    id: choose_repo
-    run: |
-      if [[ ${{ github.ref }} == *beta* ]]; then
-        echo "repo=fury" >> $GITHUB_OUTPUT
-      else
-        echo "repo=pypi" >> $GITHUB_OUTPUT
-      fi
-  - name: Publish to Fury
-    if: steps.choose_repo.outputs.repo == 'fury'
-    shell: bash
-    env:
-      FURY_TOKEN: ${{ inputs.fury_token }}
-    run: |
-      WHEEL=$(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/
-  - name: Publish to PyPI
-    if: steps.choose_repo.outputs.repo == 'pypi'
-    uses: pypa/gh-action-pypi-publish@release/v1
-    with:
-      packages-dir: target/wheels/

AGENTS.md

@@ -37,10 +37,13 @@ Before committing changes, run formatting for every language you touched. At min
   and run targeted tests through `cd python && uv run ...`.
 * TypeScript changes: run the relevant `npm`/`pnpm` lint, format, build, and docs commands in `nodejs`.
 
-Before creating a PR, make sure the PR title follows Conventional Commits, such as
-`fix: support nested field paths in native index creation` or
-`feat(python): add dataset multiprocessing support`. The semantic-release check uses the
-PR title and body as the merge commit message, so a non-conventional PR title will fail CI.
+Before creating a PR, the exact value passed to `gh pr create --title` must follow
+Conventional Commits, such as `fix: support nested field paths in native index creation`
+or `feat(python): add dataset multiprocessing support`. Do not use a plain natural
+language summary like `Support nested field paths in native index creation` as the PR
+title. The semantic-release check uses the PR title and body as the merge commit message,
+so a non-conventional PR title will fail CI. After creating a PR, read the remote PR title
+back and fix it immediately if it is not conventional.
 
 ## Coding tips
 

Cargo.toml

@@ -13,20 +13,20 @@ categories = ["database-implementations"]
 rust-version = "1.91.0"
 
 [workspace.dependencies]
-lance = { "version" = "=7.0.0-beta.13", default-features = false, "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
-lance-core = { "version" = "=7.0.0-beta.13", "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
-lance-datagen = { "version" = "=7.0.0-beta.13", "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
-lance-file = { "version" = "=7.0.0-beta.13", "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
-lance-io = { "version" = "=7.0.0-beta.13", default-features = false, "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
-lance-index = { "version" = "=7.0.0-beta.13", "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
-lance-linalg = { "version" = "=7.0.0-beta.13", "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
-lance-namespace = { "version" = "=7.0.0-beta.13", "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
-lance-namespace-impls = { "version" = "=7.0.0-beta.13", default-features = false, "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
-lance-table = { "version" = "=7.0.0-beta.13", "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
-lance-testing = { "version" = "=7.0.0-beta.13", "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
-lance-datafusion = { "version" = "=7.0.0-beta.13", "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
-lance-encoding = { "version" = "=7.0.0-beta.13", "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
-lance-arrow = { "version" = "=7.0.0-beta.13", "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
+lance = { "version" = "=7.2.0-beta.3", default-features = false, "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
+lance-core = { "version" = "=7.2.0-beta.3", "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
+lance-datagen = { "version" = "=7.2.0-beta.3", "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
+lance-file = { "version" = "=7.2.0-beta.3", "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
+lance-io = { "version" = "=7.2.0-beta.3", default-features = false, "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
+lance-index = { "version" = "=7.2.0-beta.3", "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
+lance-linalg = { "version" = "=7.2.0-beta.3", "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
+lance-namespace = { "version" = "=7.2.0-beta.3", "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
+lance-namespace-impls = { "version" = "=7.2.0-beta.3", default-features = false, "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
+lance-table = { "version" = "=7.2.0-beta.3", "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
+lance-testing = { "version" = "=7.2.0-beta.3", "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
+lance-datafusion = { "version" = "=7.2.0-beta.3", "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
+lance-encoding = { "version" = "=7.2.0-beta.3", "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
+lance-arrow = { "version" = "=7.2.0-beta.3", "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
 ahash = "0.8"
 # Note that this one does not include pyarrow
 arrow = { version = "58.0.0", optional = false }

ci/check_lance_release.py

@@ -112,25 +112,25 @@ def fetch_remote_tags() -> List[TagInfo]:
             "api",
             "-X",
             "GET",
-            f"repos/{LANCE_REPO}/git/refs/tags",
-            "--paginate",
+            f"repos/{LANCE_REPO}/releases",
             "--jq",
-            ".[].ref",
+            ".[].tag_name",
+            "-F",
+            "per_page=20",
         ]
     )
     tags: List[TagInfo] = []
     for line in output.splitlines():
-        ref = line.strip()
-        if not ref.startswith("refs/tags/v"):
+        tag = line.strip()
+        if not tag.startswith("v"):
             continue
-        tag = ref.split("refs/tags/")[-1]
         version = tag.lstrip("v")
         try:
             tags.append(TagInfo(tag=tag, version=version, semver=parse_semver(version)))
         except ValueError:
             continue
     if not tags:
-        raise RuntimeError("No Lance tags could be parsed from GitHub API output")
+        raise RuntimeError("No Lance releases could be parsed from GitHub API output")
     return tags
 
 

ci/update_lance_dependency.py

@@ -0,0 +1,126 @@
+#!/usr/bin/env python3
+"""Prepare a Lance dependency update for LanceDB."""
+
+from __future__ import annotations
+
+import argparse
+import json
+import re
+import subprocess
+import sys
+from pathlib import Path
+from typing import Sequence
+
+try:
+    from check_lance_release import parse_semver
+except ModuleNotFoundError:
+    # Supports importing as ci.update_lance_dependency from tests or ad hoc checks.
+    from ci.check_lance_release import parse_semver  # type: ignore
+
+
+def normalize_version(raw: str) -> str:
+    value = raw.strip()
+    value = value.removeprefix("refs/tags/")
+    value = value.removeprefix("v")
+    try:
+        parse_semver(value)
+    except ValueError:
+        raise ValueError(f"Unsupported Lance version or tag: {raw}")
+    return value
+
+
+def normalized_tag(version: str) -> str:
+    return f"v{version}"
+
+
+def branch_name(version: str) -> str:
+    suffix = re.sub(r"[^a-zA-Z0-9]+", "-", version).strip("-")
+    suffix = re.sub(r"-+", "-", suffix)
+    return f"codex/update-lance-{suffix}"
+
+
+def commit_type(version: str) -> str:
+    prerelease = version.split("-", maxsplit=1)[1] if "-" in version else ""
+    return "chore" if "beta" in prerelease or "rc" in prerelease else "feat"
+
+
+def metadata_for(version: str) -> dict[str, str]:
+    kind = commit_type(version)
+    message = f"{kind}: update lance dependency to v{version}"
+    return {
+        "version": version,
+        "tag": normalized_tag(version),
+        "branch_name": branch_name(version),
+        "commit_type": kind,
+        "commit_message": message,
+        "pr_title": message,
+    }
+
+
+def run_command(cmd: Sequence[str], *, cwd: Path) -> None:
+    subprocess.run(cmd, cwd=cwd, check=True)
+
+
+def update_java_lance_core_version(repo_root: Path, version: str) -> None:
+    pom_path = repo_root / "java" / "pom.xml"
+    contents = pom_path.read_text(encoding="utf-8")
+    updated, count = re.subn(
+        r"(<lance-core\.version>)[^<]+(</lance-core\.version>)",
+        rf"\g<1>{version}\g<2>",
+        contents,
+        count=1,
+    )
+    if count != 1:
+        raise RuntimeError(
+            "Expected exactly one <lance-core.version> entry in java/pom.xml"
+        )
+    pom_path.write_text(updated, encoding="utf-8")
+
+
+def write_github_outputs(path: str | None, payload: dict[str, str]) -> None:
+    if not path:
+        return
+    with open(path, "a", encoding="utf-8") as output:
+        for key, value in payload.items():
+            output.write(f"{key}={value}\n")
+
+
+def main(argv: Sequence[str] | None = None) -> int:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument(
+        "tag_or_version",
+        help="Lance tag or version, for example refs/tags/v7.2.0-beta.1 or 7.2.0",
+    )
+    parser.add_argument(
+        "--repo-root",
+        type=Path,
+        default=Path(__file__).resolve().parents[1],
+        help="Path to the lancedb repository root",
+    )
+    parser.add_argument(
+        "--github-output",
+        default=None,
+        help="Optional GitHub Actions output file to receive metadata fields",
+    )
+    parser.add_argument(
+        "--metadata-only",
+        action="store_true",
+        help="Only print derived metadata; do not modify dependency files",
+    )
+    args = parser.parse_args(argv)
+
+    repo_root = args.repo_root.resolve()
+    version = normalize_version(args.tag_or_version)
+    payload = metadata_for(version)
+
+    if not args.metadata_only:
+        run_command([sys.executable, "ci/set_lance_version.py", version], cwd=repo_root)
+        update_java_lance_core_version(repo_root, version)
+
+    write_github_outputs(args.github_output, payload)
+    print(json.dumps(payload, sort_keys=True))
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())

docs/src/java/java.md

@@ -14,7 +14,7 @@ Add the following dependency to your `pom.xml`:
 <dependency>
     <groupId>com.lancedb</groupId>
     <artifactId>lancedb-core</artifactId>
-    <version>0.29.1-beta.0</version>
+    <version>0.30.1-beta.0</version>
 </dependency>
 ```
 

docs/src/js/classes/MergeInsertBuilder.md

@@ -76,6 +76,57 @@ the query optimizer chooses a suboptimal path.
 
 ***
 
+### useLsmWrite()
+
+```ts
+useLsmWrite(useLsmWrite): MergeInsertBuilder
+```
+
+Controls whether the merge uses the MemWAL LSM write path.
+
+By default (unset), a `mergeInsert` on a table with an LSM write spec is
+routed through Lance's MemWAL shard writer, and a table without one uses
+the standard path. Pass `false` to force the standard path even when a
+spec is set. Pass `true` to require a spec — `mergeInsert` rejects if none
+is installed.
+
+#### Parameters
+
+* **useLsmWrite**: `boolean`
+    Whether to use the LSM write path.
+
+#### Returns
+
+[`MergeInsertBuilder`](MergeInsertBuilder.md)
+
+***
+
+### validateSingleShard()
+
+```ts
+validateSingleShard(validateSingleShard): MergeInsertBuilder
+```
+
+Controls how an LSM merge checks that its input targets a single shard.
+
+When a table has an LSM write spec, every row in a `mergeInsert` call must
+route to the same shard. When `true` (the default), every row is inspected
+to verify this. When `false`, only the first row is inspected and the
+shard it routes to is used for the whole input — a faster path for callers
+that have already pre-sharded their input. Has no effect on tables without
+an LSM write spec.
+
+#### Parameters
+
+* **validateSingleShard**: `boolean`
+    Whether to check every row routes to one shard. Defaults to `true`.
+
+#### Returns
+
+[`MergeInsertBuilder`](MergeInsertBuilder.md)
+
+***
+
 ### whenMatchedUpdateAll()
 
 ```ts

docs/src/js/classes/Table.md

@@ -187,6 +187,25 @@ Any attempt to use the table after it is closed will result in an error.
 
 ***
 
+### closeLsmWriters()
+
+```ts
+abstract closeLsmWriters(): Promise<void>
+```
+
+Drain and close any cached MemWAL shard writers held for this table.
+
+When an [LsmWriteSpec](../interfaces/LsmWriteSpec.md) is installed, `mergeInsert` opens MemWAL
+shard writers and caches them for reuse across calls. This closes them,
+flushing pending data; writers reopen lazily on the next `mergeInsert`.
+It is a no-op when no writers are cached.
+
+#### Returns
+
+`Promise`&lt;`void`&gt;
+
+***
+
 ### countRows()
 
 ```ts
@@ -975,6 +994,29 @@ based on the row being updated (e.g. "my_col + 1")
 
 ***
 
+### updateFieldMetadata()
+
+```ts
+abstract updateFieldMetadata(updates): Promise<UpdateFieldMetadataResult>
+```
+
+Update per-field (column) metadata.
+
+#### Parameters
+
+* **updates**: [`FieldMetadataUpdate`](../interfaces/FieldMetadataUpdate.md)[]
+    One or more per-field updates. Each
+    update's metadata is merged into the field's existing metadata by default;
+    a value of `null` deletes that key, and `replace: true` swaps the whole map.
+
+#### Returns
+
+`Promise`&lt;[`UpdateFieldMetadataResult`](../interfaces/UpdateFieldMetadataResult.md)&gt;
+
+resolves to the new table version.
+
+***
+
 ### vectorSearch()
 
 ```ts

docs/src/js/globals.md

@@ -65,6 +65,7 @@
 - [DropNamespaceOptions](interfaces/DropNamespaceOptions.md)
 - [DropNamespaceResponse](interfaces/DropNamespaceResponse.md)
 - [ExecutableQuery](interfaces/ExecutableQuery.md)
+- [FieldMetadataUpdate](interfaces/FieldMetadataUpdate.md)
 - [FragmentStatistics](interfaces/FragmentStatistics.md)
 - [FragmentSummaryStats](interfaces/FragmentSummaryStats.md)
 - [FtsOptions](interfaces/FtsOptions.md)
@@ -101,6 +102,7 @@
 - [TimeoutConfig](interfaces/TimeoutConfig.md)
 - [TlsConfig](interfaces/TlsConfig.md)
 - [TokenResponse](interfaces/TokenResponse.md)
+- [UpdateFieldMetadataResult](interfaces/UpdateFieldMetadataResult.md)
 - [UpdateOptions](interfaces/UpdateOptions.md)
 - [UpdateResult](interfaces/UpdateResult.md)
 - [Version](interfaces/Version.md)

docs/src/js/interfaces/ConnectionOptions.md

@@ -70,16 +70,20 @@ client used by manifest-enabled native connections.
 optional readConsistencyInterval: number;
 ```
 
-(For LanceDB OSS only): The interval, in seconds, at which to check for
-updates to the table from other processes. If None, then consistency is not
-checked. For performance reasons, this is the default. For strong
-consistency, set this to zero seconds. Then every read will check for
-updates from other processes. As a compromise, you can set this to a
-non-zero value for eventual consistency. If more than that interval
-has passed since the last check, then the table will be checked for updates.
-Note: this consistency only applies to read operations. Write operations are
+The interval, in seconds, at which to check for updates to the table
+from other processes. If None, then consistency is not checked. For
+performance reasons, this is the default. For strong consistency, set
+this to zero seconds. Then every read will check for updates from other
+processes. As a compromise, you can set this to a non-zero value for
+eventual consistency. If more than that interval has passed since the
+last check, then the table will be checked for updates. Note: this
+consistency only applies to read operations. Write operations are
 always consistent.
 
+Stronger consistency is not free. The smaller the interval, the more
+often each read pays the cost of checking for updates against object
+storage, raising per-read latency and cost.
+
 ***
 
 ### region?

docs/src/js/interfaces/FieldMetadataUpdate.md

@@ -0,0 +1,41 @@
+[**@lancedb/lancedb**](../README.md) • **Docs**
+
+***
+
+[@lancedb/lancedb](../globals.md) / FieldMetadataUpdate
+
+# Interface: FieldMetadataUpdate
+
+A per-field metadata update, addressed by dot-path.
+
+## Properties
+
+### metadata
+
+```ts
+metadata: Record<string, null | string>;
+```
+
+Metadata key/value pairs. Merged into the field's existing metadata by
+default; a value of `null` deletes that key.
+
+***
+
+### path
+
+```ts
+path: string;
+```
+
+Dot-separated path to the field. For a top-level column this is just its
+name; for a nested field it's the path, e.g. "a.b.c".
+
+***
+
+### replace?
+
+```ts
+optional replace: boolean;
+```
+
+If true, replace the field's entire metadata map instead of merging.

docs/src/js/interfaces/LsmWriteSpec.md

@@ -11,7 +11,10 @@ Specification selecting Lance's MemWAL LSM-style write path for
 
 `specType` is `"bucket"`, `"identity"`, or `"unsharded"`. For `"bucket"`,
 `column` and `numBuckets` are required; for `"identity"`, `column` is
-required.
+required and must be a deterministic function of the unenforced primary
+key (every row with a given primary key must always produce the same
+`column` value, or upserts of that key can land in different shards and a
+stale version can win).
 
 ## Properties
 

docs/src/js/interfaces/MergeResult.md

@@ -32,6 +32,14 @@ numInsertedRows: number;
 
 ***
 
+### numRows
+
+```ts
+numRows: number;
+```
+
+***
+
 ### numUpdatedRows
 
 ```ts

docs/src/js/interfaces/UpdateFieldMetadataResult.md

@@ -0,0 +1,15 @@
+[**@lancedb/lancedb**](../README.md) • **Docs**
+
+***
+
+[@lancedb/lancedb](../globals.md) / UpdateFieldMetadataResult
+
+# Interface: UpdateFieldMetadataResult
+
+## Properties
+
+### version
+
+```ts
+version: number;
+```

java/lancedb-core/pom.xml

@@ -8,7 +8,7 @@
     <parent>
       <groupId>com.lancedb</groupId>
       <artifactId>lancedb-parent</artifactId>
-      <version>0.29.1-beta.0</version>
+      <version>0.30.1-beta.0</version>
       <relativePath>../pom.xml</relativePath>
     </parent>
 

java/pom.xml

@@ -6,7 +6,7 @@
 
     <groupId>com.lancedb</groupId>
     <artifactId>lancedb-parent</artifactId>
-    <version>0.29.1-beta.0</version>
+    <version>0.30.1-beta.0</version>
     <packaging>pom</packaging>
     <name>${project.artifactId}</name>
     <description>LanceDB Java SDK Parent POM</description>
@@ -28,7 +28,7 @@
     <properties>
         <project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
         <arrow.version>15.0.0</arrow.version>
-        <lance-core.version>7.0.0-beta.13</lance-core.version>
+        <lance-core.version>7.2.0-beta.1</lance-core.version>
         <spotless.skip>false</spotless.skip>
         <spotless.version>2.30.0</spotless.version>
         <spotless.java.googlejavaformat.version>1.7</spotless.java.googlejavaformat.version>

nodejs/Cargo.toml

@@ -1,7 +1,7 @@
 [package]
 name = "lancedb-nodejs"
 edition.workspace = true
-version = "0.29.1-beta.0"
+version = "0.30.1-beta.0"
 publish = false
 license.workspace = true
 description.workspace = true

nodejs/__test__/connection.test.ts

@@ -171,18 +171,22 @@ describe("given a connection", () => {
 
     let manifestDir =
       tmpDir.name + "/test_manifest_paths_v2_empty.lance/_versions";
-    readdirSync(manifestDir).forEach((file) => {
-      expect(file).toMatch(/^\d{20}\.manifest$/);
-    });
+    readdirSync(manifestDir)
+      .filter((f) => f.endsWith(".manifest"))
+      .forEach((file) => {
+        expect(file).toMatch(/^\d{20}\.manifest$/);
+      });
 
     table = (await db.createTable("test_manifest_paths_v2", [{ id: 1 }], {
       enableV2ManifestPaths: true,
     })) as LocalTable;
     expect(await table.usesV2ManifestPaths()).toBe(true);
     manifestDir = tmpDir.name + "/test_manifest_paths_v2.lance/_versions";
-    readdirSync(manifestDir).forEach((file) => {
-      expect(file).toMatch(/^\d{20}\.manifest$/);
-    });
+    readdirSync(manifestDir)
+      .filter((f) => f.endsWith(".manifest"))
+      .forEach((file) => {
+        expect(file).toMatch(/^\d{20}\.manifest$/);
+      });
   });
 
   it("should be able to migrate tables to the V2 manifest paths", async () => {
@@ -199,16 +203,20 @@ describe("given a connection", () => {
 
     const manifestDir =
       tmpDir.name + "/test_manifest_path_migration.lance/_versions";
-    readdirSync(manifestDir).forEach((file) => {
-      expect(file).toMatch(/^\d\.manifest$/);
-    });
+    readdirSync(manifestDir)
+      .filter((f) => f.endsWith(".manifest"))
+      .forEach((file) => {
+        expect(file).toMatch(/^\d\.manifest$/);
+      });
 
     await table.migrateManifestPathsV2();
     expect(await table.usesV2ManifestPaths()).toBe(true);
 
-    readdirSync(manifestDir).forEach((file) => {
-      expect(file).toMatch(/^\d{20}\.manifest$/);
-    });
+    readdirSync(manifestDir)
+      .filter((f) => f.endsWith(".manifest"))
+      .forEach((file) => {
+        expect(file).toMatch(/^\d{20}\.manifest$/);
+      });
   });
 });
 

nodejs/__test__/table.test.ts

@@ -28,6 +28,7 @@ import {
   List,
   Schema,
   SchemaLike,
+  Struct,
   Type,
   Uint8,
   Utf8,
@@ -780,6 +781,113 @@ describe("When creating an index", () => {
     expect(indices2.length).toBe(0);
   });
 
+  it("should create and search a nested vector index", async () => {
+    const db = await connect(tmpDir.name);
+    const nestedSchema = new Schema([
+      new Field("id", new Int32(), true),
+      new Field(
+        "image",
+        new Struct([
+          new Field(
+            "embedding",
+            new FixedSizeList(2, new Field("item", new Float32(), true)),
+            true,
+          ),
+        ]),
+        true,
+      ),
+    ]);
+    const nestedTable = await db.createTable(
+      "nested_vector",
+      makeArrowTable(
+        Array.from({ length: 300 }, (_, id) => ({
+          id,
+          image: { embedding: [id, id + 1] },
+        })),
+        { schema: nestedSchema },
+      ),
+    );
+
+    await nestedTable.createIndex("image.embedding", {
+      name: "image_embedding_idx",
+    });
+    const indices = await nestedTable.listIndices();
+    expect(indices).toContainEqual({
+      name: "image_embedding_idx",
+      indexType: "IvfPq",
+      columns: ["image.embedding"],
+    });
+
+    const explicit = await nestedTable
+      .query()
+      .nearestTo([0.0, 1.0])
+      .column("image.embedding")
+      .limit(1)
+      .toArray();
+    const inferred = await nestedTable
+      .query()
+      .nearestTo([0.0, 1.0])
+      .limit(1)
+      .toArray();
+    expect(inferred[0].id).toEqual(explicit[0].id);
+  });
+
+  it("should report multiple nested vector candidates", async () => {
+    const db = await connect(tmpDir.name);
+    const nestedSchema = new Schema([
+      new Field(
+        "image",
+        new Struct([
+          new Field(
+            "embedding",
+            new FixedSizeList(2, new Field("item", new Float32(), true)),
+            true,
+          ),
+        ]),
+        true,
+      ),
+      new Field(
+        "text",
+        new Struct([
+          new Field(
+            "embedding",
+            new FixedSizeList(2, new Field("item", new Float32(), true)),
+            true,
+          ),
+        ]),
+        true,
+      ),
+    ]);
+    const nestedTable = await db.createTable(
+      "multiple_nested_vectors",
+      makeArrowTable(
+        [
+          {
+            image: { embedding: [0.0, 1.0] },
+            text: { embedding: [2.0, 3.0] },
+          },
+        ],
+        { schema: nestedSchema },
+      ),
+    );
+
+    await expect(
+      nestedTable.query().nearestTo([0.0, 1.0]).limit(1).toArray(),
+    ).rejects.toThrow(/image\.embedding.*text\.embedding/);
+  });
+
+  it("should report when no default vector column exists", async () => {
+    const db = await connect(tmpDir.name);
+    const noVectorTable = await db.createTable(
+      "no_vector",
+      makeArrowTable([{ id: 0, label: "cat" }]),
+    );
+
+    await expect(
+      noVectorTable.query().nearestTo([0.0, 1.0]).limit(1).toArray(),
+    ).rejects.toThrow(/No vector column/);
+  });
+
   it("should wait for index readiness", async () => {
     // Create an index and then wait for it to be ready
     await tbl.createIndex("vec");
@@ -1463,6 +1571,33 @@ describe("schema evolution", function () {
     expect(await table.schema()).toEqual(expectedSchema3);
   });
 
+  it("can update field metadata", async function () {
+    const con = await connect(tmpDir.name);
+    const table = await con.createTable("fm", [
+      { id: 1, category: "a" },
+      { id: 2, category: "b" },
+    ]);
+
+    const res = await table.updateFieldMetadata([
+      { path: "category", metadata: { unit: "label", pii: "false" } },
+    ]);
+    expect(res).toHaveProperty("version");
+    expect(res.version).toBe(2);
+
+    let cat = (await table.schema()).fields.find((f) => f.name === "category");
+    expect(cat?.metadata.get("unit")).toBe("label");
+    expect(cat?.metadata.get("pii")).toBe("false");
+
+    // merge: add a key, delete one via null, keep the rest
+    await table.updateFieldMetadata([
+      { path: "category", metadata: { source: "import", pii: null } },
+    ]);
+    cat = (await table.schema()).fields.find((f) => f.name === "category");
+    expect(cat?.metadata.get("unit")).toBe("label"); // preserved
+    expect(cat?.metadata.get("source")).toBe("import"); // added
+    expect(cat?.metadata.has("pii")).toBe(false); // deleted
+  });
+
   it("can cast to various types", async function () {
     const con = await connect(tmpDir.name);
 
@@ -2517,3 +2652,97 @@ describe("setLsmWriteSpec / unsetLsmWriteSpec", () => {
     ).rejects.toThrow();
   });
 });
+
+describe("LSM merge insert", () => {
+  let tmpDir: tmp.DirResult;
+
+  beforeEach(() => {
+    tmpDir = tmp.dirSync({ unsafeCleanup: true });
+  });
+  afterEach(() => tmpDir.removeCallback());
+
+  async function bucketTable(conn: Connection): Promise<Table> {
+    // The primary key column must be non-nullable.
+    const table = await conn.createEmptyTable(
+      "t",
+      new arrow.Schema([
+        new arrow.Field("id", new arrow.Utf8(), false),
+        new arrow.Field("value", new arrow.Float64(), true),
+      ]),
+    );
+    await table.add([
+      { id: "a", value: 1 },
+      { id: "b", value: 2 },
+    ]);
+    await table.setUnenforcedPrimaryKey("id");
+    // numBuckets = 1: every row routes to the single bucket.
+    await table.setLsmWriteSpec({
+      specType: "bucket",
+      column: "id",
+      numBuckets: 1,
+    });
+    return table;
+  }
+
+  it("routes merge_insert through the shard writer", async () => {
+    const conn = await connect(tmpDir.name);
+    const table = await bucketTable(conn);
+
+    const res = await table
+      .mergeInsert("id")
+      .whenMatchedUpdateAll()
+      .whenNotMatchedInsertAll()
+      .execute([
+        { id: "c", value: 3 },
+        { id: "d", value: 4 },
+      ]);
+    // LSM path: rows go to the MemWAL, so only numRows is populated.
+    expect(res.numRows).toBe(2);
+    expect(res.version).toBe(0);
+    expect(res.numInsertedRows).toBe(0);
+
+    await table.closeLsmWriters();
+  });
+
+  it("falls back to the standard path with useLsmWrite(false)", async () => {
+    const conn = await connect(tmpDir.name);
+    const table = await bucketTable(conn);
+
+    const res = await table
+      .mergeInsert("id")
+      .whenNotMatchedInsertAll()
+      .useLsmWrite(false)
+      .execute([
+        { id: "b", value: 9 },
+        { id: "e", value: 5 },
+      ]);
+    // Standard path commits: id="e" inserted ("b" already exists).
+    expect(res.numInsertedRows).toBe(1);
+    expect(await table.countRows()).toBe(3);
+  });
+
+  it("supports validateSingleShard(false)", async () => {
+    const conn = await connect(tmpDir.name);
+    const table = await bucketTable(conn);
+
+    const res = await table
+      .mergeInsert("id")
+      .whenMatchedUpdateAll()
+      .whenNotMatchedInsertAll()
+      .validateSingleShard(false)
+      .execute([{ id: "f", value: 6 }]);
+    expect(res.numRows).toBe(1);
+  });
+
+  it("rejects a non-upsert merge under an LSM spec", async () => {
+    const conn = await connect(tmpDir.name);
+    const table = await bucketTable(conn);
+
+    await expect(
+      table
+        .mergeInsert("id")
+        .whenNotMatchedInsertAll()
+        .execute([{ id: "g", value: 7 }]),
+    ).rejects.toThrow();
+  });
+});

nodejs/lancedb/index.ts

@@ -42,6 +42,7 @@ export {
   AddResult,
   AddColumnsResult,
   AlterColumnsResult,
+  UpdateFieldMetadataResult,
   DeleteResult,
   DropColumnsResult,
   UpdateResult,
@@ -117,6 +118,7 @@ export {
   WriteProgress,
   LsmWriteSpec,
   ColumnAlteration,
+  FieldMetadataUpdate,
 } from "./table";
 
 export {

nodejs/lancedb/merge.ts

@@ -87,6 +87,41 @@ export class MergeInsertBuilder {
       this.#schema,
     );
   }
+  /**
+   * Controls whether the merge uses the MemWAL LSM write path.
+   *
+   * By default (unset), a `mergeInsert` on a table with an LSM write spec is
+   * routed through Lance's MemWAL shard writer, and a table without one uses
+   * the standard path. Pass `false` to force the standard path even when a
+   * spec is set. Pass `true` to require a spec — `mergeInsert` rejects if none
+   * is installed.
+   *
+   * @param useLsmWrite - Whether to use the LSM write path.
+   */
+  useLsmWrite(useLsmWrite: boolean): MergeInsertBuilder {
+    return new MergeInsertBuilder(
+      this.#native.useLsmWrite(useLsmWrite),
+      this.#schema,
+    );
+  }
+  /**
+   * Controls how an LSM merge checks that its input targets a single shard.
+   *
+   * When a table has an LSM write spec, every row in a `mergeInsert` call must
+   * route to the same shard. When `true` (the default), every row is inspected
+   * to verify this. When `false`, only the first row is inspected and the
+   * shard it routes to is used for the whole input — a faster path for callers
+   * that have already pre-sharded their input. Has no effect on tables without
+   * an LSM write spec.
+   *
+   * @param validateSingleShard - Whether to check every row routes to one shard. Defaults to `true`.
+   */
+  validateSingleShard(validateSingleShard: boolean): MergeInsertBuilder {
+    return new MergeInsertBuilder(
+      this.#native.validateSingleShard(validateSingleShard),
+      this.#schema,
+    );
+  }
   /**
    * Executes the merge insert operation
    *

nodejs/lancedb/table.ts

@@ -32,6 +32,7 @@ import {
   OptimizeStats,
   TableStatistics,
   Tags,
+  UpdateFieldMetadataResult,
   UpdateResult,
   Table as _NativeTable,
 } from "./native";
@@ -161,7 +162,10 @@ export interface Version {
  *
  * `specType` is `"bucket"`, `"identity"`, or `"unsharded"`. For `"bucket"`,
  * `column` and `numBuckets` are required; for `"identity"`, `column` is
- * required.
+ * required and must be a deterministic function of the unenforced primary
+ * key (every row with a given primary key must always produce the same
+ * `column` value, or upserts of that key can land in different shards and a
+ * stale version can win).
  */
 export interface LsmWriteSpec {
   /** One of `"bucket"`, `"identity"`, or `"unsharded"`. */
@@ -505,6 +509,18 @@ export abstract class Table {
   abstract alterColumns(
     columnAlterations: ColumnAlteration[],
   ): Promise<AlterColumnsResult>;
+
+  /**
+   * Update per-field (column) metadata.
+   * @param {FieldMetadataUpdate[]} updates One or more per-field updates. Each
+   * update's metadata is merged into the field's existing metadata by default;
+   * a value of `null` deletes that key, and `replace: true` swaps the whole map.
+   * @returns {Promise<UpdateFieldMetadataResult>} resolves to the new table version.
+   */
+  abstract updateFieldMetadata(
+    updates: FieldMetadataUpdate[],
+  ): Promise<UpdateFieldMetadataResult>;
+
   /**
    * Drop one or more columns from the dataset
    *
@@ -567,6 +583,16 @@ export abstract class Table {
    * @returns {Promise<void>}
    */
   abstract unsetLsmWriteSpec(): Promise<void>;
+  /**
+   * Drain and close any cached MemWAL shard writers held for this table.
+   *
+   * When an {@link LsmWriteSpec} is installed, `mergeInsert` opens MemWAL
+   * shard writers and caches them for reuse across calls. This closes them,
+   * flushing pending data; writers reopen lazily on the next `mergeInsert`.
+   * It is a no-op when no writers are cached.
+   * @returns {Promise<void>}
+   */
+  abstract closeLsmWriters(): Promise<void>;
   /** Retrieve the version of the table */
 
   abstract version(): Promise<number>;
@@ -1024,6 +1050,12 @@ export class LocalTable extends Table {
     return await this.inner.alterColumns(processedAlterations);
   }
 
+  async updateFieldMetadata(
+    updates: FieldMetadataUpdate[],
+  ): Promise<UpdateFieldMetadataResult> {
+    return await this.inner.updateFieldMetadata(updates);
+  }
+
   async dropColumns(columnNames: string[]): Promise<DropColumnsResult> {
     return await this.inner.dropColumns(columnNames);
   }
@@ -1041,6 +1073,10 @@ export class LocalTable extends Table {
     return await this.inner.unsetLsmWriteSpec();
   }
 
+  async closeLsmWriters(): Promise<void> {
+    return await this.inner.closeLsmWriters();
+  }
+
   async version(): Promise<number> {
     return await this.inner.version();
   }
@@ -1186,3 +1222,19 @@ export interface ColumnAlteration {
   /** Set the new nullability. Note that a nullable column cannot be made non-nullable. */
   nullable?: boolean;
 }
+
+/** A per-field metadata update, addressed by dot-path. */
+export interface FieldMetadataUpdate {
+  /**
+   * Dot-separated path to the field. For a top-level column this is just its
+   * name; for a nested field it's the path, e.g. "a.b.c".
+   */
+  path: string;
+  /**
+   * Metadata key/value pairs. Merged into the field's existing metadata by
+   * default; a value of `null` deletes that key.
+   */
+  metadata: Record<string, string | null>;
+  /** If true, replace the field's entire metadata map instead of merging. */
+  replace?: boolean;
+}

nodejs/npm/darwin-arm64/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-darwin-arm64",
-	"version": "0.29.1-beta.0",
+	"version": "0.30.1-beta.0",
 	"os": ["darwin"],
 	"cpu": ["arm64"],
 	"main": "lancedb.darwin-arm64.node",

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

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-arm64-gnu",
-	"version": "0.29.1-beta.0",
+	"version": "0.30.1-beta.0",
 	"os": ["linux"],
 	"cpu": ["arm64"],
 	"main": "lancedb.linux-arm64-gnu.node",

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

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-arm64-musl",
-	"version": "0.29.1-beta.0",
+	"version": "0.30.1-beta.0",
 	"os": ["linux"],
 	"cpu": ["arm64"],
 	"main": "lancedb.linux-arm64-musl.node",

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

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-x64-gnu",
-	"version": "0.29.1-beta.0",
+	"version": "0.30.1-beta.0",
 	"os": ["linux"],
 	"cpu": ["x64"],
 	"main": "lancedb.linux-x64-gnu.node",

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

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-x64-musl",
-	"version": "0.29.1-beta.0",
+	"version": "0.30.1-beta.0",
 	"os": ["linux"],
 	"cpu": ["x64"],
 	"main": "lancedb.linux-x64-musl.node",

nodejs/npm/win32-arm64-msvc/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lancedb/lancedb-win32-arm64-msvc",
-  "version": "0.29.1-beta.0",
+  "version": "0.30.1-beta.0",
   "os": [
     "win32"
   ],

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

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-win32-x64-msvc",
-	"version": "0.29.1-beta.0",
+	"version": "0.30.1-beta.0",
 	"os": ["win32"],
 	"cpu": ["x64"],
 	"main": "lancedb.win32-x64-msvc.node",

nodejs/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "@lancedb/lancedb",
-  "version": "0.29.1-beta.0",
+  "version": "0.30.1-beta.0",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "@lancedb/lancedb",
-      "version": "0.29.1-beta.0",
+      "version": "0.30.1-beta.0",
       "cpu": [
         "x64",
         "arm64"

nodejs/package.json

@@ -11,7 +11,7 @@
     "ann"
   ],
   "private": false,
-  "version": "0.29.1-beta.0",
+  "version": "0.30.1-beta.0",
   "main": "dist/index.js",
   "exports": {
     ".": "./dist/index.js",

nodejs/src/lib.rs

@@ -24,15 +24,19 @@ mod util;
 #[napi(object)]
 #[derive(Debug)]
 pub struct ConnectionOptions {
-    /// (For LanceDB OSS only): The interval, in seconds, at which to check for
-    /// updates to the table from other processes. If None, then consistency is not
-    /// checked. For performance reasons, this is the default. For strong
-    /// consistency, set this to zero seconds. Then every read will check for
-    /// updates from other processes. As a compromise, you can set this to a
-    /// non-zero value for eventual consistency. If more than that interval
-    /// has passed since the last check, then the table will be checked for updates.
-    /// Note: this consistency only applies to read operations. Write operations are
+    /// The interval, in seconds, at which to check for updates to the table
+    /// from other processes. If None, then consistency is not checked. For
+    /// performance reasons, this is the default. For strong consistency, set
+    /// this to zero seconds. Then every read will check for updates from other
+    /// processes. As a compromise, you can set this to a non-zero value for
+    /// eventual consistency. If more than that interval has passed since the
+    /// last check, then the table will be checked for updates. Note: this
+    /// consistency only applies to read operations. Write operations are
     /// always consistent.
+    ///
+    /// Stronger consistency is not free. The smaller the interval, the more
+    /// often each read pays the cost of checking for updates against object
+    /// storage, raising per-read latency and cost.
     pub read_consistency_interval: Option<f64>,
     /// (For LanceDB OSS only): configuration for object storage.
     ///

nodejs/src/merge.rs

@@ -50,6 +50,20 @@ impl NativeMergeInsertBuilder {
         this
     }
 
+    #[napi]
+    pub fn use_lsm_write(&self, use_lsm_write: bool) -> Self {
+        let mut this = self.clone();
+        this.inner.use_lsm_write(use_lsm_write);
+        this
+    }
+
+    #[napi]
+    pub fn validate_single_shard(&self, validate_single_shard: bool) -> Self {
+        let mut this = self.clone();
+        this.inner.validate_single_shard(validate_single_shard);
+        this
+    }
+
     #[napi(catch_unwind)]
     pub async fn execute(&self, buf: Buffer) -> napi::Result<MergeResult> {
         let data = ipc_file_to_batches(buf.to_vec())

nodejs/src/table.rs

@@ -5,8 +5,9 @@ use std::collections::HashMap;
 
 use lancedb::ipc::{ipc_file_to_batches, ipc_file_to_schema};
 use lancedb::table::{
-    AddDataMode, ColumnAlteration as LanceColumnAlteration, Duration, NewColumnTransform,
-    OptimizeAction, OptimizeOptions, Table as LanceDbTable,
+    AddDataMode, ColumnAlteration as LanceColumnAlteration, Duration,
+    FieldMetadataUpdate as LanceFieldMetadataUpdate, NewColumnTransform, OptimizeAction,
+    OptimizeOptions, Table as LanceDbTable,
 };
 use napi::bindgen_prelude::*;
 use napi::threadsafe_function::{ThreadsafeFunction, ThreadsafeFunctionCallMode};
@@ -355,6 +356,23 @@ impl Table {
         Ok(res.into())
     }
 
+    #[napi(catch_unwind)]
+    pub async fn update_field_metadata(
+        &self,
+        updates: Vec<FieldMetadataUpdate>,
+    ) -> napi::Result<UpdateFieldMetadataResult> {
+        let updates = updates
+            .into_iter()
+            .map(LanceFieldMetadataUpdate::from)
+            .collect::<Vec<_>>();
+        let res = self
+            .inner_ref()?
+            .update_field_metadata(&updates)
+            .await
+            .default_error()?;
+        Ok(res.into())
+    }
+
     #[napi(catch_unwind)]
     pub async fn drop_columns(&self, columns: Vec<String>) -> napi::Result<DropColumnsResult> {
         let col_refs = columns.iter().map(String::as_str).collect::<Vec<_>>();
@@ -391,6 +409,11 @@ impl Table {
             .default_error()
     }
 
+    #[napi(catch_unwind)]
+    pub async fn close_lsm_writers(&self) -> napi::Result<()> {
+        self.inner_ref()?.close_lsm_writers().await.default_error()
+    }
+
     #[napi(catch_unwind)]
     pub async fn version(&self) -> napi::Result<i64> {
         self.inner_ref()?
@@ -742,6 +765,29 @@ pub struct ColumnAlteration {
     pub nullable: Option<bool>,
 }
 
+/// A per-field metadata update, addressed by dot-path. Merges into the field's
+/// existing metadata by default; a `null` value deletes a key, and `replace`
+/// swaps the field's entire metadata map.
+#[napi(object)]
+pub struct FieldMetadataUpdate {
+    /// Dot-separated path to the field (e.g. "embedding" or "a.b.c").
+    pub path: String,
+    /// Metadata keys to set; a `null` value deletes that key.
+    pub metadata: HashMap<String, Option<String>>,
+    /// If true, replace the field's entire metadata map instead of merging.
+    pub replace: Option<bool>,
+}
+
+impl From<FieldMetadataUpdate> for LanceFieldMetadataUpdate {
+    fn from(js: FieldMetadataUpdate) -> Self {
+        Self {
+            path: js.path,
+            metadata: js.metadata,
+            replace: js.replace.unwrap_or(false),
+        }
+    }
+}
+
 impl TryFrom<ColumnAlteration> for LanceColumnAlteration {
     type Error = String;
     fn try_from(js: ColumnAlteration) -> std::result::Result<Self, Self::Error> {
@@ -940,6 +986,7 @@ pub struct MergeResult {
     pub num_updated_rows: i64,
     pub num_deleted_rows: i64,
     pub num_attempts: i64,
+    pub num_rows: i64,
 }
 
 impl From<lancedb::table::MergeResult> for MergeResult {
@@ -950,6 +997,7 @@ impl From<lancedb::table::MergeResult> for MergeResult {
             num_updated_rows: value.num_updated_rows as i64,
             num_deleted_rows: value.num_deleted_rows as i64,
             num_attempts: value.num_attempts as i64,
+            num_rows: value.num_rows as i64,
         }
     }
 }
@@ -980,6 +1028,19 @@ impl From<lancedb::table::AlterColumnsResult> for AlterColumnsResult {
     }
 }
 
+#[napi(object)]
+pub struct UpdateFieldMetadataResult {
+    pub version: i64,
+}
+
+impl From<lancedb::table::UpdateFieldMetadataResult> for UpdateFieldMetadataResult {
+    fn from(value: lancedb::table::UpdateFieldMetadataResult) -> Self {
+        Self {
+            version: value.version as i64,
+        }
+    }
+}
+
 #[napi(object)]
 pub struct DropColumnsResult {
     pub version: i64,

python/.bumpversion.toml

@@ -1,5 +1,5 @@
 [tool.bumpversion]
-current_version = "0.32.1-beta.0"
+current_version = "0.33.1-beta.0"
 parse = """(?x)
     (?P<major>0|[1-9]\\d*)\\.
     (?P<minor>0|[1-9]\\d*)\\.

python/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "lancedb-python"
-version = "0.32.1-beta.0"
+version = "0.33.1-beta.0"
 publish = false
 edition.workspace = true
 description = "Python bindings for LanceDB"

python/python/lancedb/__init__.py

@@ -94,7 +94,6 @@ def connect(
     host_override: str, optional
         The override url for LanceDB Cloud.
     read_consistency_interval: timedelta, default None
-        (For LanceDB OSS only)
         The interval at which to check for updates to the table from other
         processes. If None, then consistency is not checked. For performance
         reasons, this is the default. For strong consistency, set this to
@@ -104,6 +103,10 @@ def connect(
         the last check, then the table will be checked for updates. Note: this
         consistency only applies to read operations. Write operations are
         always consistent.
+
+        Stronger consistency is not free. The smaller the interval, the more
+        often each read pays the cost of checking for updates against object
+        storage, raising per-read latency and cost.
     client_config: ClientConfig or dict, optional
         Configuration options for the LanceDB Cloud HTTP client. If a dict, then
         the keys are the attributes of the ClientConfig class. If None, then the
@@ -147,6 +150,13 @@ def connect(
     >>> db = lancedb.connect("s3://my-bucket/lancedb",
     ...                      storage_options={"aws_access_key_id": "***"})
 
+    For tests and temporary data, use an in-memory database:
+
+    >>> db = lancedb.connect("memory://")
+
+    In-memory databases are not persisted. Tables are dropped when the last
+    connection or table handle referencing them is closed.
+
     Connect to LanceDB cloud:
 
     >>> db = lancedb.connect("db://my_database", api_key="ldb_...",
@@ -210,6 +220,7 @@ def connect(
             request_thread_pool=request_thread_pool,
             client_config=client_config,
             storage_options=storage_options,
+            read_consistency_interval=read_consistency_interval,
             **kwargs,
         )
     _check_s3_bucket_with_dots(str(uri), storage_options)
@@ -304,6 +315,15 @@ def deserialize_conn(
             manifest_enabled=parsed.get("manifest_enabled", False),
             namespace_client_properties=parsed.get("namespace_client_properties"),
         )
+    elif connection_type == "remote":
+        return RemoteDBConnection(
+            parsed["db_url"],
+            parsed["api_key"],
+            parsed.get("region", "us-east-1"),
+            host_override=parsed.get("host_override"),
+            client_config=parsed.get("client_config"),
+            storage_options=storage_options,
+        )
     else:
         raise ValueError(f"Unknown connection_type: {connection_type}")
 
@@ -336,7 +356,6 @@ async def connect_async(
     host_override: str, optional
         The override url for LanceDB Cloud.
     read_consistency_interval: timedelta, default None
-        (For LanceDB OSS only)
         The interval at which to check for updates to the table from other
         processes. If None, then consistency is not checked. For performance
         reasons, this is the default. For strong consistency, set this to
@@ -346,6 +365,10 @@ async def connect_async(
         the last check, then the table will be checked for updates. Note: this
         consistency only applies to read operations. Write operations are
         always consistent.
+
+        Stronger consistency is not free. The smaller the interval, the more
+        often each read pays the cost of checking for updates against object
+        storage, raising per-read latency and cost.
     client_config: ClientConfig or dict, optional
         Configuration options for the LanceDB Cloud HTTP client. If a dict, then
         the keys are the attributes of the ClientConfig class. If None, then the
@@ -378,6 +401,8 @@ async def connect_async(
     ...     db = await lancedb.connect_async("s3://my-bucket/lancedb",
     ...                                      storage_options={
     ...                                          "aws_access_key_id": "***"})
+    ...     # For tests and temporary data, use an in-memory database
+    ...     db = await lancedb.connect_async("memory://")
     ...     # Connect to LanceDB cloud
     ...     db = await lancedb.connect_async("db://my_database", api_key="ldb_...",
     ...                                      client_config={

python/python/lancedb/_lancedb.pyi

@@ -208,6 +208,9 @@ class Table:
     async def alter_columns(
         self, columns: list[dict[str, Any]]
     ) -> AlterColumnsResult: ...
+    async def update_field_metadata(
+        self, updates: list[dict[str, Any]]
+    ) -> UpdateFieldMetadataResult: ...
     async def optimize(
         self,
         *,
@@ -220,6 +223,7 @@ class Table:
     async def set_unenforced_primary_key(self, columns: List[str]) -> None: ...
     async def set_lsm_write_spec(self, spec: LsmWriteSpec) -> None: ...
     async def unset_lsm_write_spec(self) -> None: ...
+    async def close_lsm_writers(self) -> None: ...
     @property
     def tags(self) -> Tags: ...
     def query(self) -> Query: ...
@@ -420,6 +424,7 @@ class MergeResult:
     num_inserted_rows: int
     num_deleted_rows: int
     num_attempts: int
+    num_rows: int
 
 class LsmWriteSpec:
     """Specification selecting Lance's MemWAL LSM-style write path for
@@ -458,6 +463,9 @@ class AddColumnsResult:
 class AlterColumnsResult:
     version: int
 
+class UpdateFieldMetadataResult:
+    version: int
+
 class DropColumnsResult:
     version: int
 

python/python/lancedb/db.py

@@ -8,7 +8,17 @@ from abc import abstractmethod
 from datetime import timedelta
 from pathlib import Path
 import sys
-from typing import TYPE_CHECKING, Any, Dict, Iterable, List, Literal, Optional, Union
+from typing import (
+    TYPE_CHECKING,
+    Any,
+    Dict,
+    Generator,
+    Iterable,
+    List,
+    Literal,
+    Optional,
+    Union,
+)
 
 if sys.version_info >= (3, 12):
     from typing import override
@@ -313,7 +323,7 @@ class DBConnection(EnforceOverrides):
         >>> data = [{"vector": [1.1, 1.2], "lat": 45.5, "long": -122.7},
         ...         {"vector": [0.2, 1.8], "lat": 40.1, "long":  -74.1}]
         >>> db.create_table("my_table", data)
-        LanceTable(name='my_table', version=1, ...)
+        LanceTable(name='my_table', ...)
         >>> db["my_table"].head()
         pyarrow.Table
         vector: fixed_size_list<item: float>[2]
@@ -334,7 +344,7 @@ class DBConnection(EnforceOverrides):
         ...    "long": [-122.7, -74.1]
         ... })
         >>> db.create_table("table2", data)
-        LanceTable(name='table2', version=1, ...)
+        LanceTable(name='table2', ...)
         >>> db["table2"].head()
         pyarrow.Table
         vector: fixed_size_list<item: float>[2]
@@ -357,7 +367,7 @@ class DBConnection(EnforceOverrides):
         ...   pa.field("long", pa.float32())
         ... ])
         >>> db.create_table("table3", data, schema = custom_schema)
-        LanceTable(name='table3', version=1, ...)
+        LanceTable(name='table3', ...)
         >>> db["table3"].head()
         pyarrow.Table
         vector: fixed_size_list<item: float>[2]
@@ -391,7 +401,7 @@ class DBConnection(EnforceOverrides):
         ...     pa.field("price", pa.float32()),
         ... ])
         >>> db.create_table("table4", make_batches(), schema=schema)
-        LanceTable(name='table4', version=1, ...)
+        LanceTable(name='table4', ...)
 
         """
         raise NotImplementedError
@@ -568,15 +578,15 @@ class LanceDBConnection(DBConnection):
     >>> db = lancedb.connect("./.lancedb")
     >>> db.create_table("my_table", data=[{"vector": [1.1, 1.2], "b": 2},
     ...                                   {"vector": [0.5, 1.3], "b": 4}])
-    LanceTable(name='my_table', version=1, ...)
+    LanceTable(name='my_table', ...)
     >>> db.create_table("another_table", data=[{"vector": [0.4, 0.4], "b": 6}])
-    LanceTable(name='another_table', version=1, ...)
+    LanceTable(name='another_table', ...)
     >>> sorted(db.table_names())
     ['another_table', 'my_table']
     >>> len(db)
     2
     >>> db["my_table"]
-    LanceTable(name='my_table', version=1, ...)
+    LanceTable(name='my_table', ...)
     >>> "my_table" in db
     True
     >>> db.drop_table("my_table")
@@ -847,11 +857,20 @@ class LanceDBConnection(DBConnection):
             )
         )
 
+    def _all_table_names(self) -> Generator[str, None, None]:
+        page_token = None
+        while True:
+            response = self.list_tables(page_token=page_token)
+            yield from response.tables
+            page_token = response.page_token
+            if not page_token:
+                return
+
     def __len__(self) -> int:
-        return len(self.table_names())
+        return sum(1 for _ in self._all_table_names())
 
     def __contains__(self, name: str) -> bool:
-        return name in self.table_names()
+        return name in self._all_table_names()
 
     @override
     def create_table(

python/python/lancedb/index.py

@@ -281,6 +281,9 @@ class HnswPq:
     m: int = 20
     ef_construction: int = 300
     target_partition_size: Optional[int] = None
+    # Name of the accelerator (e.g. "cuda") to use for IVF training. When set,
+    # create_index() dispatches to pylance to build the index on the accelerator.
+    accelerator: Optional[str] = None
 
 
 @dataclass
@@ -386,6 +389,9 @@ class HnswSq:
     m: int = 20
     ef_construction: int = 300
     target_partition_size: Optional[int] = None
+    # Name of the accelerator (e.g. "cuda") to use for IVF training. When set,
+    # create_index() dispatches to pylance to build the index on the accelerator.
+    accelerator: Optional[str] = None
 
 
 @dataclass
@@ -579,6 +585,9 @@ class IvfFlat:
     max_iterations: int = 50
     sample_rate: int = 256
     target_partition_size: Optional[int] = None
+    # Name of the accelerator (e.g. "cuda") to use for IVF training. When set,
+    # create_index() dispatches to pylance to build the index on the accelerator.
+    accelerator: Optional[str] = None
 
 
 @dataclass
@@ -609,6 +618,9 @@ class IvfSq:
     max_iterations: int = 50
     sample_rate: int = 256
     target_partition_size: Optional[int] = None
+    # Name of the accelerator (e.g. "cuda") to use for IVF training. When set,
+    # create_index() dispatches to pylance to build the index on the accelerator.
+    accelerator: Optional[str] = None
 
 
 @dataclass
@@ -739,6 +751,9 @@ class IvfPq:
     max_iterations: int = 50
     sample_rate: int = 256
     target_partition_size: Optional[int] = None
+    # Name of the accelerator (e.g. "cuda") to use for IVF training. When set,
+    # create_index() dispatches to pylance to build the index on the accelerator.
+    accelerator: Optional[str] = None
 
 
 @dataclass
@@ -792,6 +807,9 @@ class IvfRq:
     max_iterations: int = 50
     sample_rate: int = 256
     target_partition_size: Optional[int] = None
+    # Name of the accelerator (e.g. "cuda") to use for IVF training. When set,
+    # create_index() dispatches to pylance to build the index on the accelerator.
+    accelerator: Optional[str] = None
 
 
 __all__ = [

python/python/lancedb/merge.py

@@ -34,6 +34,8 @@ class LanceMergeInsertBuilder(object):
         self._when_not_matched_by_source_condition = None
         self._timeout = None
         self._use_index = True
+        self._use_lsm_write = None
+        self._validate_single_shard = None
 
     def when_matched_update_all(
         self, *, where: Optional[str] = None
@@ -96,6 +98,46 @@ class LanceMergeInsertBuilder(object):
         self._use_index = use_index
         return self
 
+    def use_lsm_write(self, use_lsm_write: bool) -> LanceMergeInsertBuilder:
+        """
+        Controls whether the merge uses the MemWAL LSM write path.
+
+        By default (unset), a `merge_insert` on a table with an LSM write spec
+        is routed through Lance's MemWAL shard writer, and a table without one
+        uses the standard path. Pass `False` to force the standard path even
+        when a spec is set. Pass `True` to require a spec — `merge_insert`
+        raises an error if none is installed.
+
+        Parameters
+        ----------
+        use_lsm_write: bool
+            Whether to use the LSM write path.
+        """
+        self._use_lsm_write = use_lsm_write
+        return self
+
+    def validate_single_shard(
+        self, validate_single_shard: bool
+    ) -> LanceMergeInsertBuilder:
+        """
+        Controls how an LSM merge checks that its input targets a single shard.
+
+        When a table has an LSM write spec, every row in a `merge_insert` call
+        must route to the same shard. When `True` (the default), every row is
+        inspected to verify this. When `False`, only the first row is inspected
+        and the shard it routes to is used for the whole input — a faster path
+        for callers that have already pre-sharded their input.
+
+        Has no effect on tables without an LSM write spec.
+
+        Parameters
+        ----------
+        validate_single_shard: bool
+            Whether to check every row routes to one shard. Defaults to `True`.
+        """
+        self._validate_single_shard = validate_single_shard
+        return self
+
     def execute(
         self,
         new_data: DATA,

python/python/lancedb/permutation.py

@@ -3,12 +3,13 @@
 
 import copy
 import json
+import os
 
 from deprecation import deprecated
 import pyarrow as pa
 
 from ._lancedb import async_permutation_builder, PermutationReader
-from .table import LanceTable
+from .table import LanceTable, Table
 from .background_loop import LOOP
 from .util import batch_to_tensor, batch_to_tensor_rows
 from typing import Any, Callable, Iterator, Literal, Optional, TYPE_CHECKING, Union
@@ -354,6 +355,49 @@ class Transforms:
 DEFAULT_BATCH_SIZE = 100
 
 
+def _table_to_pickle_state(table: Table) -> dict[str, Any]:
+    from .remote.table import RemoteTable
+
+    if isinstance(table, RemoteTable):
+        return {
+            "kind": "remote",
+            "table": table,
+        }
+
+    if not isinstance(table, LanceTable):
+        raise ValueError(f"Cannot pickle table of type {type(table)!r}")
+
+    base_uri = table._conn.uri
+    if base_uri.startswith("memory://"):
+        return {
+            "kind": "memory",
+            "name": table.name,
+            "data": table.to_arrow(),
+        }
+
+    return {
+        "kind": "local",
+        "name": table.name,
+        "uri": base_uri,
+        "namespace": table._namespace_path,
+        "storage_options": table._conn.storage_options,
+    }
+
+
+def _table_from_pickle_state(state: dict[str, Any]) -> Table:
+    from . import connect
+
+    kind = state["kind"]
+    if kind == "remote":
+        return state["table"]
+    if kind == "memory":
+        return connect("memory://").create_table(state["name"], state["data"])
+    if kind == "local":
+        db = connect(state["uri"], storage_options=state["storage_options"])
+        return db.open_table(state["name"], namespace_path=state["namespace"] or None)
+    raise ValueError(f"Unknown table pickle state kind: {kind}")
+
+
 class Permutation:
     """
     A Permutation is a view of a dataset that can be used as input to model training
@@ -369,15 +413,15 @@ class Permutation:
 
     def __init__(
         self,
-        base_table: LanceTable,
-        permutation_table: Optional[LanceTable],
+        base_table: Table,
+        permutation_table: Optional[Table],
         split: int,
         selection: dict[str, str],
         batch_size: int,
         transform_fn: Callable[pa.RecordBatch, Any],
         offset: Optional[int] = None,
         limit: Optional[int] = None,
-        connection_factory: Optional[Callable[[str], LanceTable]] = None,
+        connection_factory: Optional[Callable[[str], Table]] = None,
         _reader: Optional[PermutationReader] = None,
     ):
         """
@@ -397,6 +441,7 @@ class Permutation:
         if _reader is None:
             _reader = LOOP.run(self._build_reader())
         self.reader: PermutationReader = _reader
+        self._pid = os.getpid()
 
     async def _build_reader(self) -> PermutationReader:
         reader = await PermutationReader.from_tables(
@@ -428,29 +473,25 @@ class Permutation:
         return new
 
     def with_connection_factory(
-        self, connection_factory: Callable[[str], LanceTable]
+        self, connection_factory: Callable[[str], Table]
     ) -> "Permutation":
         """
         Creates a new permutation that will use ``connection_factory`` to reopen
         the base table when this permutation is unpickled in a worker process.
 
-        The factory is a callable that takes a single argument — the base table
-        name — and returns a [LanceTable]. It must be picklable; the worker
+        The factory is a callable that takes a single argument, the base table
+        name, and returns a LanceDB table. It must be picklable; the worker
         will pickle it via standard ``pickle`` and call it to recover the base
         table. Picklable callables in practice means top-level (module-level)
         functions, ``functools.partial`` of such functions, or instances of
         picklable classes implementing ``__call__``. Lambdas and closures over
         local variables don't pickle with the default protocol.
 
-        Setting a factory is necessary when the URI alone is not enough to
-        re-open the connection — most importantly for LanceDB Cloud (``db://``)
-        connections, where ``api_key`` and ``region`` aren't recoverable from
-        the connection object after construction.
-
-        For local file or cloud-storage paths the factory is optional: if not
-        set, ``__getstate__`` falls back to capturing
-        ``(uri, storage_options, namespace_path)`` and re-opening via
-        ``lancedb.connect(uri, storage_options=...)``.
+        A factory is optional for normal local and remote LanceDB connections:
+        if not set, ``__getstate__`` captures the table's own picklable reopen
+        state. Use a factory when that default state is not enough, for example
+        when credentials should be loaded from the worker environment instead
+        of being embedded in the pickle.
 
         Examples
         --------
@@ -508,7 +549,7 @@ class Permutation:
         return new
 
     @classmethod
-    def identity(cls, table: LanceTable) -> "Permutation":
+    def identity(cls, table: Table) -> "Permutation":
         """
         Creates an identity permutation for the given table.
         """
@@ -517,8 +558,8 @@ class Permutation:
     @classmethod
     def from_tables(
         cls,
-        base_table: LanceTable,
-        permutation_table: Optional[LanceTable] = None,
+        base_table: Table,
+        permutation_table: Optional[Table] = None,
         split: Optional[Union[str, int]] = None,
     ) -> "Permutation":
         """
@@ -594,11 +635,10 @@ class Permutation:
 
         The base table is captured either via a user-supplied
         ``connection_factory`` (see [with_connection_factory]) or, as a
-        fallback, by introspecting ``(uri, storage_options, namespace_path)``
-        on the connection. The permutation table — always an in-memory
-        LanceDB table — is captured as a pyarrow Table (which pickles via
-        Arrow IPC natively). The reader is dropped from the wire format;
-        ``__setstate__`` rebuilds it from the restored tables.
+        fallback, by the table's own picklable reopen state. The permutation
+        table is captured as a pyarrow Table (which pickles via Arrow IPC
+        natively). The reader is dropped from the wire format and rebuilt
+        lazily on first use.
         """
         permutation_data: Optional[pa.Table] = None
         if self.permutation_table is not None:
@@ -622,39 +662,9 @@ class Permutation:
             # namespace from the existing connection.
             return common
 
-        # URI-introspection fallback: only viable for native (OSS) connections
-        # where (uri, storage_options) is enough to reopen. Remote / cloud
-        # connections don't expose recoverable api_key / region — those users
-        # must call with_connection_factory().
-        try:
-            base_uri = self.base_table._conn.uri
-            storage_options = self.base_table._conn.storage_options
-        except AttributeError as e:
-            raise ValueError(
-                "Cannot pickle this Permutation: the base table's connection "
-                "does not expose a uri/storage_options, which usually means it "
-                "is a remote (LanceDB Cloud) connection. Call "
-                "Permutation.with_connection_factory(...) first to provide a "
-                "picklable callable that re-opens the base table from a worker "
-                "process."
-            ) from e
-
-        if base_uri.startswith("memory://"):
-            # In-memory base tables don't exist in any worker process by
-            # default, so dump the entire base table into the pickle. This
-            # can be expensive for large datasets — users with large
-            # in-memory base tables should either persist them or set a
-            # connection_factory.
-            return {
-                **common,
-                "base_table_data": self.base_table.to_arrow(),
-            }
-
         return {
             **common,
-            "base_table_uri": base_uri,
-            "base_table_namespace": self.base_table._namespace_path,
-            "base_table_storage_options": storage_options,
+            "base_table_state": _table_to_pickle_state(self.base_table),
         }
 
     def __setstate__(self, state: dict[str, Any]) -> None:
@@ -663,6 +673,8 @@ class Permutation:
         connection_factory = state["connection_factory"]
         if connection_factory is not None:
             base_table = connection_factory(state["base_table_name"])
+        elif "base_table_state" in state:
+            base_table = _table_from_pickle_state(state["base_table_state"])
         elif "base_table_data" in state:
             # In-memory base table inlined into the pickle; rebuild the same
             # way we rebuild the in-memory permutation table.
@@ -680,7 +692,7 @@ class Permutation:
                 namespace_path=state["base_table_namespace"] or None,
             )
 
-        permutation_table: Optional[LanceTable] = None
+        permutation_table: Optional[Table] = None
         if state["permutation_data"] is not None:
             mem_db = connect("memory://")
             permutation_table = mem_db.create_table(
@@ -696,10 +708,28 @@ class Permutation:
         self.offset = state["offset"]
         self.limit = state["limit"]
         self.connection_factory = connection_factory
+        self.reader = None
+        self._pid = None
+
+    def _ensure_open(self) -> None:
+        pid = os.getpid()
+        if self.reader is not None and getattr(self, "_pid", None) == pid:
+            return
+        # The reader owns Rust-side table handles. Rebuild it after unpickle or
+        # fork even though the Python table wrappers reopen themselves.
+        if hasattr(self.base_table, "_ensure_open"):
+            self.base_table._ensure_open()
+        if self.permutation_table is not None and hasattr(
+            self.permutation_table, "_ensure_open"
+        ):
+            self.permutation_table._ensure_open()
         self.reader = LOOP.run(self._build_reader())
+        self._pid = pid
 
     @property
     def schema(self) -> pa.Schema:
+        self._ensure_open()
+
         async def do_output_schema():
             return await self.reader.output_schema(self.selection)
 
@@ -717,6 +747,7 @@ class Permutation:
         """
         The number of rows in the permutation
         """
+        self._ensure_open()
         return self.reader.count_rows()
 
     @property
@@ -875,6 +906,7 @@ class Permutation:
         If skip_last_batch is True, the last batch will be skipped if it is not a
         multiple of batch_size.
         """
+        self._ensure_open()
 
         async def get_iter():
             return await self.reader.read(self.selection, batch_size=batch_size)
@@ -976,6 +1008,7 @@ class Permutation:
         so `with_format` and `with_transform` affect this method in the same way
         they affect iteration.
         """
+        self._ensure_open()
 
         async def do_take_offsets():
             return await self.reader.take_offsets(offsets, selection=self.selection)
@@ -1011,9 +1044,11 @@ class Permutation:
         """
         Skip the first `skip` rows of the permutation
         """
+        self._ensure_open()
         new = copy.copy(self)
         new.offset = skip
         new.reader = LOOP.run(new._build_reader())
+        new._pid = os.getpid()
         return new
 
     @deprecated(details="Use with_take instead")
@@ -1032,9 +1067,11 @@ class Permutation:
         """
         Limit the permutation to `limit` rows (following any `skip`)
         """
+        self._ensure_open()
         new = copy.copy(self)
         new.limit = limit
         new.reader = LOOP.run(new._build_reader())
+        new._pid = os.getpid()
         return new
 
     @deprecated(details="Use with_repeat instead")

python/python/lancedb/query.py

@@ -3,12 +3,14 @@
 
 from __future__ import annotations
 
+import asyncio
 from abc import ABC, abstractmethod
 from concurrent.futures import ThreadPoolExecutor
-from enum import Enum
 from datetime import timedelta
+from enum import Enum
 from typing import (
     TYPE_CHECKING,
+    Any,
     Dict,
     List,
     Literal,
@@ -17,41 +19,40 @@ from typing import (
     Type,
     TypeVar,
     Union,
-    Any,
 )
 
-import asyncio
 import deprecation
 import numpy as np
 import pyarrow as pa
 import pyarrow.compute as pc
 import pydantic
+from typing_extensions import Annotated
 
-from lancedb.pydantic import PYDANTIC_VERSION
+from lancedb._lancedb import fts_query_to_json
 from lancedb.background_loop import LOOP
+from lancedb.pydantic import PYDANTIC_VERSION
 
 from . import __version__
 from .arrow import AsyncRecordBatchReader
 from .dependencies import pandas as pd
+from .expr import Expr
 from .rerankers.base import Reranker
 from .rerankers.rrf import RRFReranker
 from .rerankers.util import check_reranker_result
 from .util import flatten_columns
-from .expr import Expr
-from lancedb._lancedb import fts_query_to_json
-from typing_extensions import Annotated
 
 if TYPE_CHECKING:
     import sys
+
     import PIL
     import polars as pl
 
-    from ._lancedb import Query as LanceQuery
     from ._lancedb import FTSQuery as LanceFTSQuery
     from ._lancedb import HybridQuery as LanceHybridQuery
-    from ._lancedb import VectorQuery as LanceVectorQuery
-    from ._lancedb import TakeQuery as LanceTakeQuery
     from ._lancedb import PyQueryRequest
+    from ._lancedb import Query as LanceQuery
+    from ._lancedb import TakeQuery as LanceTakeQuery
+    from ._lancedb import VectorQuery as LanceVectorQuery
     from .common import VEC
     from .pydantic import LanceModel
     from .table import Table
@@ -3348,16 +3349,18 @@ class BaseQueryBuilder(object):
             If not specified, no timeout is applied. If the query does not
             complete within the specified time, an error will be raised.
         """
-        async_iter = LOOP.run(self._inner.execute(max_batch_length, timeout))
+        async_reader = LOOP.run(
+            self._inner.to_batches(max_batch_length=max_batch_length, timeout=timeout)
+        )
 
         def iter_sync():
             try:
                 while True:
-                    yield LOOP.run(async_iter.__anext__())
+                    yield LOOP.run(async_reader.__anext__())
             except StopAsyncIteration:
                 return
 
-        return pa.RecordBatchReader.from_batches(async_iter.schema, iter_sync())
+        return pa.RecordBatchReader.from_batches(async_reader.schema, iter_sync())
 
     def to_arrow(self, timeout: Optional[timedelta] = None) -> pa.Table:
         """

python/python/lancedb/remote/db.py

@@ -3,6 +3,7 @@
 
 
 from datetime import timedelta
+import json
 import logging
 from concurrent.futures import ThreadPoolExecutor
 import sys
@@ -17,7 +18,7 @@ else:
 
 # Remove this import to fix circular dependency
 # from lancedb import connect_async
-from lancedb.remote import ClientConfig
+from lancedb.remote import ClientConfig, RetryConfig, TimeoutConfig, TlsConfig
 import pyarrow as pa
 
 from ..common import DATA
@@ -36,6 +37,64 @@ from ..table import Table
 from ..util import validate_table_name
 
 
+def _duration_seconds(value: Optional[timedelta]) -> Optional[float]:
+    return value.total_seconds() if value is not None else None
+
+
+def _timeout_config_to_dict(
+    config: Optional[TimeoutConfig],
+) -> Optional[dict[str, Any]]:
+    if config is None:
+        return None
+    return {
+        "timeout": _duration_seconds(config.timeout),
+        "connect_timeout": _duration_seconds(config.connect_timeout),
+        "read_timeout": _duration_seconds(config.read_timeout),
+        "pool_idle_timeout": _duration_seconds(config.pool_idle_timeout),
+    }
+
+
+def _retry_config_to_dict(config: RetryConfig) -> dict[str, Any]:
+    return {
+        "retries": config.retries,
+        "connect_retries": config.connect_retries,
+        "read_retries": config.read_retries,
+        "backoff_factor": config.backoff_factor,
+        "backoff_jitter": config.backoff_jitter,
+        "statuses": config.statuses,
+    }
+
+
+def _tls_config_to_dict(config: Optional[TlsConfig]) -> Optional[dict[str, Any]]:
+    if config is None:
+        return None
+    return {
+        "cert_file": config.cert_file,
+        "key_file": config.key_file,
+        "ssl_ca_cert": config.ssl_ca_cert,
+        "assert_hostname": config.assert_hostname,
+    }
+
+
+def _client_config_to_dict(config: ClientConfig) -> dict[str, Any]:
+    if config.header_provider is not None:
+        raise ValueError(
+            "Cannot serialize a remote connection with a header_provider. "
+            "Use static api_key/extra_headers or provide a worker-side "
+            "connection factory instead."
+        )
+    return {
+        "user_agent": config.user_agent,
+        "retry_config": _retry_config_to_dict(config.retry_config),
+        "timeout_config": _timeout_config_to_dict(config.timeout_config),
+        "extra_headers": config.extra_headers,
+        "id_delimiter": config.id_delimiter,
+        "tls_config": _tls_config_to_dict(config.tls_config),
+        "header_provider": None,
+        "user_id": config.user_id,
+    }
+
+
 class RemoteDBConnection(DBConnection):
     """A connection to a remote LanceDB database."""
 
@@ -50,6 +109,7 @@ class RemoteDBConnection(DBConnection):
         connection_timeout: Optional[float] = None,
         read_timeout: Optional[float] = None,
         storage_options: Optional[Dict[str, str]] = None,
+        read_consistency_interval: Optional[timedelta] = None,
     ):
         """Connect to a remote LanceDB database."""
         if isinstance(client_config, dict):
@@ -88,6 +148,11 @@ class RemoteDBConnection(DBConnection):
         parsed = urlparse(db_url)
         if parsed.scheme != "db":
             raise ValueError(f"Invalid scheme: {parsed.scheme}, only accepts db://")
+        self.db_url = db_url
+        self.api_key = api_key
+        self.region = region
+        self.host_override = host_override
+        self.storage_options = storage_options
         self.db_name = parsed.netloc
 
         self.client_config = client_config
@@ -103,12 +168,27 @@ class RemoteDBConnection(DBConnection):
                 host_override=host_override,
                 client_config=client_config,
                 storage_options=storage_options,
+                read_consistency_interval=read_consistency_interval,
             )
         )
 
     def __repr__(self) -> str:
         return f"RemoteConnect(name={self.db_name})"
 
+    @override
+    def serialize(self) -> str:
+        return json.dumps(
+            {
+                "connection_type": "remote",
+                "db_url": self.db_url,
+                "api_key": self.api_key,
+                "region": self.region,
+                "host_override": self.host_override,
+                "client_config": _client_config_to_dict(self.client_config),
+                "storage_options": self.storage_options,
+            }
+        )
+
     @override
     def list_namespaces(
         self,
@@ -329,7 +409,12 @@ class RemoteDBConnection(DBConnection):
             )
 
         table = LOOP.run(self._conn.open_table(name, namespace_path=namespace_path))
-        return RemoteTable(table, self.db_name)
+        return RemoteTable(
+            table,
+            self.db_name,
+            connection_state=self.serialize,
+            namespace_path=namespace_path,
+        )
 
     def clone_table(
         self,
@@ -378,7 +463,12 @@ class RemoteDBConnection(DBConnection):
                 is_shallow=is_shallow,
             )
         )
-        return RemoteTable(table, self.db_name)
+        return RemoteTable(
+            table,
+            self.db_name,
+            connection_state=self.serialize,
+            namespace_path=target_namespace_path,
+        )
 
     @override
     def create_table(
@@ -523,7 +613,12 @@ class RemoteDBConnection(DBConnection):
                 fill_value=fill_value,
             )
         )
-        return RemoteTable(table, self.db_name)
+        return RemoteTable(
+            table,
+            self.db_name,
+            connection_state=self.serialize,
+            namespace_path=namespace_path,
+        )
 
     @override
     def drop_table(self, name: str, namespace_path: Optional[List[str]] = None):

python/python/lancedb/remote/table.py

@@ -2,15 +2,30 @@
 # SPDX-FileCopyrightText: Copyright The LanceDB Authors
 
 from datetime import timedelta
+import deprecation
 import logging
 from functools import cached_property
-from typing import Any, Callable, Dict, Iterable, List, Optional, Union, Literal
+import os
+from typing import (
+    Any,
+    Callable,
+    Dict,
+    Iterable,
+    List,
+    Optional,
+    Union,
+    Literal,
+    overload,
+)
 import warnings
 
+from lancedb import __version__
+
 from lancedb._lancedb import (
     AddColumnsResult,
     AddResult,
     AlterColumnsResult,
+    UpdateFieldMetadataResult,
     DeleteResult,
     DropColumnsResult,
     IndexConfig,
@@ -32,6 +47,7 @@ from lancedb.index import (
     LabelList,
 )
 from lancedb.remote.db import LOOP
+from lancedb.table import IndexConfigType, KNOWN_METRICS
 import pyarrow as pa
 
 from lancedb.common import DATA, VEC, VECTOR_COLUMN_NAME
@@ -49,14 +65,80 @@ class RemoteTable(Table):
         self,
         table: AsyncTable,
         db_name: str,
+        *,
+        connection_state: Optional[Union[str, Callable[[], str]]] = None,
+        namespace_path: Optional[List[str]] = None,
     ):
-        self._table = table
+        self._table_handle = table
+        self._name = table.name
         self.db_name = db_name
+        self._connection_state = connection_state
+        self._namespace_path = list(namespace_path or [])
+        self._checkout_version: Optional[int] = None
+        self._pid = os.getpid()
+
+    def _serialized_connection_state(self) -> str:
+        if self._connection_state is None:
+            raise RuntimeError(
+                "Cannot reopen this remote table because it does not carry "
+                "serialized connection state"
+            )
+        if callable(self._connection_state):
+            self._connection_state = self._connection_state()
+        return self._connection_state
+
+    @property
+    def _table(self) -> AsyncTable:
+        self._ensure_open()
+        assert self._table_handle is not None
+        return self._table_handle
+
+    @_table.setter
+    def _table(self, table: AsyncTable) -> None:
+        self._table_handle = table
+        self._name = table.name
+        self._pid = os.getpid()
+
+    def _ensure_open(self) -> None:
+        pid = os.getpid()
+        if self._table_handle is not None and self._pid == pid:
+            return
+
+        # Pickle clears the handle; fork inherits a handle created in the
+        # parent process. In both cases reopen before touching the Rust client.
+        from lancedb import deserialize_conn
+
+        db = deserialize_conn(self._serialized_connection_state(), for_worker=True)
+        table = db.open_table(self._name, namespace_path=self._namespace_path)
+        if self._checkout_version is not None:
+            table.checkout(self._checkout_version)
+
+        self._table_handle = table._table
+        self.db_name = table.db_name
+        self._pid = pid
+
+    def __getstate__(self) -> dict:
+        return {
+            "connection_state": self._serialized_connection_state(),
+            "db_name": self.db_name,
+            "name": self.name,
+            "namespace_path": self._namespace_path,
+            "checkout_version": self._checkout_version,
+        }
+
+    def __setstate__(self, state: dict) -> None:
+        self._table_handle = None
+        self._name = state["name"]
+        self.db_name = state["db_name"]
+        self._connection_state = state["connection_state"]
+        self._namespace_path = state["namespace_path"]
+        self._checkout_version = state["checkout_version"]
+        self._pid = None
 
     @property
     def name(self) -> str:
         """The name of the table"""
-        return self._table.name
+        return self._name
 
     def __repr__(self) -> str:
         return f"RemoteTable({self.db_name}.{self.name})"
@@ -106,13 +188,19 @@ class RemoteTable(Table):
         raise NotImplementedError("to_pandas() is not yet supported on LanceDB cloud.")
 
     def checkout(self, version: Union[int, str]):
-        return LOOP.run(self._table.checkout(version))
+        result = LOOP.run(self._table.checkout(version))
+        self._checkout_version = self.version
+        return result
 
     def checkout_latest(self):
-        return LOOP.run(self._table.checkout_latest())
+        result = LOOP.run(self._table.checkout_latest())
+        self._checkout_version = None
+        return result
 
     def restore(self, version: Optional[Union[int, str]] = None):
-        return LOOP.run(self._table.restore(version))
+        result = LOOP.run(self._table.restore(version))
+        self._checkout_version = None
+        return result
 
     def list_indices(self) -> Iterable[IndexConfig]:
         """List all the indices on the table"""
@@ -122,6 +210,11 @@ class RemoteTable(Table):
         """List all the stats of a specified index"""
         return LOOP.run(self._table.index_stats(index_uuid))
 
+    @deprecation.deprecated(
+        deprecated_in="0.25.0",
+        current_version=__version__,
+        details="Use create_index() with config=BTree()/Bitmap()/LabelList() instead.",
+    )
     def create_scalar_index(
         self,
         column: str,
@@ -131,7 +224,12 @@ class RemoteTable(Table):
         wait_timeout: Optional[timedelta] = None,
         name: Optional[str] = None,
     ):
-        """Creates a scalar index
+        """Creates a scalar index.
+
+        .. deprecated:: 0.25.0
+            Use :meth:`create_index` with a BTree, Bitmap, or LabelList config instead.
+            Example: ``table.create_index("column", config=BTree())``
+
         Parameters
         ----------
         column : str
@@ -162,6 +260,11 @@ class RemoteTable(Table):
             )
         )
 
+    @deprecation.deprecated(
+        deprecated_in="0.25.0",
+        current_version=__version__,
+        details="Use create_index() with config=FTS() instead.",
+    )
     def create_fts_index(
         self,
         column: str,
@@ -182,6 +285,12 @@ class RemoteTable(Table):
         prefix_only: bool = False,
         name: Optional[str] = None,
     ):
+        """Create a full-text search index on a column.
+
+        .. deprecated:: 0.25.0
+            Use :meth:`create_index` with an FTS config instead.
+            Example: ``table.create_index("text_column", config=FTS())``
+        """
         config = FTS(
             with_position=with_position,
             base_tokenizer=base_tokenizer,
@@ -205,9 +314,43 @@ class RemoteTable(Table):
             )
         )
 
+    # New unified API overload
+    @overload
     def create_index(
         self,
-        metric="l2",
+        column: str,
+        /,
+        *,
+        config: IndexConfigType,
+        wait_timeout: Optional[timedelta] = ...,
+        name: Optional[str] = ...,
+        train: bool = ...,
+    ) -> None: ...
+
+    # Legacy API overload (deprecated)
+    @overload
+    def create_index(
+        self,
+        metric: Literal["l2", "cosine", "dot", "hamming"] = ...,
+        vector_column_name: str = ...,
+        index_cache_size: Optional[int] = ...,
+        num_partitions: Optional[int] = ...,
+        num_sub_vectors: Optional[int] = ...,
+        replace: Optional[bool] = ...,
+        accelerator: Optional[str] = ...,
+        index_type: Literal[
+            "VECTOR", "IVF_FLAT", "IVF_SQ", "IVF_PQ", "IVF_HNSW_SQ", "IVF_HNSW_PQ"
+        ] = ...,
+        wait_timeout: Optional[timedelta] = ...,
+        *,
+        num_bits: int = ...,
+        name: Optional[str] = ...,
+        train: bool = ...,
+    ) -> None: ...
+
+    def create_index(
+        self,
+        metric: str = "l2",
         vector_column_name: str = VECTOR_COLUMN_NAME,
         index_cache_size: Optional[int] = None,
         num_partitions: Optional[int] = None,
@@ -218,89 +361,113 @@ class RemoteTable(Table):
         wait_timeout: Optional[timedelta] = None,
         *,
         num_bits: int = 8,
+        config: Optional[IndexConfigType] = None,
         name: Optional[str] = None,
         train: bool = True,
     ):
-        """Create an index on the table.
+        """Create an index on a column.
 
-        Parameters
-        ----------
-        metric : str
-            The metric to use for the index. Default is "l2".
-        vector_column_name : str
-            The name of the vector column. Default is "vector".
+        This method supports both the new unified API and the legacy API
+        for backwards compatibility. The new API takes the column name as the
+        first positional argument and an index configuration object via
+        ``config``; the legacy API takes the distance metric as the first
+        argument plus separate ``vector_column_name`` / ``num_partitions`` /
+        etc. parameters, and emits a ``DeprecationWarning``.
 
         Examples
         --------
-        >>> import lancedb
-        >>> import uuid
-        >>> from lancedb.schema import vector
-        >>> db = lancedb.connect("db://...", api_key="...", # doctest: +SKIP
-        ...                      region="...") # doctest: +SKIP
-        >>> table_name = uuid.uuid4().hex
-        >>> schema = pa.schema(
-        ...     [
-        ...             pa.field("id", pa.uint32(), False),
-        ...            pa.field("vector", vector(128), False),
-        ...             pa.field("s", pa.string(), False),
-        ...     ]
+        New API (recommended):
+
+        >>> table.create_index(  # doctest: +SKIP
+        ...     "vector", config=IvfPq(distance_type="l2")
         ... )
-        >>> table = db.create_table( # doctest: +SKIP
-        ...     table_name, # doctest: +SKIP
-        ...     schema=schema, # doctest: +SKIP
+        >>> table.create_index("category", config=BTree())  # doctest: +SKIP
+        >>> table.create_index("content", config=FTS())  # doctest: +SKIP
+
+        Legacy API (deprecated):
+
+        >>> table.create_index(  # doctest: +SKIP
+        ...     "l2", vector_column_name="vector"
         ... )
-        >>> table.create_index("l2", "vector") # doctest: +SKIP
         """
+        # Detect whether this is a legacy API call
+        is_legacy = self._is_legacy_create_index_call(
+            metric,
+            config,
+            num_partitions,
+            num_sub_vectors,
+            vector_column_name,
+            accelerator,
+            index_cache_size,
+            replace,
+        )
 
-        if accelerator is not None:
-            logging.warning(
-                "GPU accelerator is not yet supported on LanceDB cloud."
-                "If you have 100M+ vectors to index,"
-                "please contact us at contact@lancedb.com"
-            )
-        if replace is not None:
-            logging.warning(
-                "replace is not supported on LanceDB cloud."
-                "Existing indexes will always be replaced."
+        if is_legacy:
+            warnings.warn(
+                "The create_index() API with metric/num_partitions parameters is "
+                "deprecated and will be removed in a future version. "
+                "Please migrate to the new unified API:\n"
+                "  # Old (deprecated):\n"
+                "  table.create_index('l2', vector_column_name='my_vector')\n"
+                "  # New (recommended):\n"
+                "  table.create_index('my_vector', config=IvfPq(distance_type='l2'))",
+                DeprecationWarning,
+                stacklevel=2,
             )
 
-        index_type = index_type.upper()
-        if index_type == "VECTOR" or index_type == "IVF_PQ":
-            config = IvfPq(
-                distance_type=metric,
-                num_partitions=num_partitions,
-                num_sub_vectors=num_sub_vectors,
-                num_bits=num_bits,
-            )
-        elif index_type == "IVF_RQ":
-            config = IvfRq(
-                distance_type=metric,
-                num_partitions=num_partitions,
-                num_bits=num_bits,
-            )
-        elif index_type == "IVF_SQ":
-            config = IvfSq(distance_type=metric, num_partitions=num_partitions)
-        elif index_type == "IVF_HNSW_PQ":
-            raise ValueError(
-                "IVF_HNSW_PQ is not supported on LanceDB cloud."
-                "Please use IVF_HNSW_SQ instead."
-            )
-        elif index_type == "IVF_HNSW_SQ":
-            config = HnswSq(distance_type=metric, num_partitions=num_partitions)
-        elif index_type == "IVF_HNSW_FLAT":
-            config = HnswFlat(distance_type=metric, num_partitions=num_partitions)
-        elif index_type == "IVF_FLAT":
-            config = IvfFlat(distance_type=metric, num_partitions=num_partitions)
+            column = vector_column_name
+
+            if accelerator is not None:
+                logging.warning(
+                    "GPU accelerator is not yet supported on LanceDB cloud."
+                    "If you have 100M+ vectors to index,"
+                    "please contact us at contact@lancedb.com"
+                )
+            if replace is not None:
+                logging.warning(
+                    "replace is not supported on LanceDB cloud."
+                    "Existing indexes will always be replaced."
+                )
+
+            idx_type = index_type.upper()
+            if idx_type == "VECTOR" or idx_type == "IVF_PQ":
+                config = IvfPq(
+                    distance_type=metric,
+                    num_partitions=num_partitions,
+                    num_sub_vectors=num_sub_vectors,
+                    num_bits=num_bits,
+                )
+            elif idx_type == "IVF_RQ":
+                config = IvfRq(
+                    distance_type=metric,
+                    num_partitions=num_partitions,
+                    num_bits=num_bits,
+                )
+            elif idx_type == "IVF_SQ":
+                config = IvfSq(distance_type=metric, num_partitions=num_partitions)
+            elif idx_type == "IVF_HNSW_PQ":
+                raise ValueError(
+                    "IVF_HNSW_PQ is not supported on LanceDB cloud."
+                    "Please use IVF_HNSW_SQ instead."
+                )
+            elif idx_type == "IVF_HNSW_SQ":
+                config = HnswSq(distance_type=metric, num_partitions=num_partitions)
+            elif idx_type == "IVF_HNSW_FLAT":
+                config = HnswFlat(distance_type=metric, num_partitions=num_partitions)
+            elif idx_type == "IVF_FLAT":
+                config = IvfFlat(distance_type=metric, num_partitions=num_partitions)
+            else:
+                raise ValueError(
+                    f"Unknown vector index type: {idx_type}. Valid options are"
+                    " 'IVF_FLAT', 'IVF_PQ', 'IVF_RQ', 'IVF_SQ',"
+                    " 'IVF_HNSW_PQ', 'IVF_HNSW_SQ', 'IVF_HNSW_FLAT'"
+                )
         else:
-            raise ValueError(
-                f"Unknown vector index type: {index_type}. Valid options are"
-                " 'IVF_FLAT', 'IVF_PQ', 'IVF_RQ', 'IVF_SQ',"
-                " 'IVF_HNSW_PQ', 'IVF_HNSW_SQ', 'IVF_HNSW_FLAT'"
-            )
+            column = metric
 
         LOOP.run(
             self._table.create_index(
-                vector_column_name,
+                column,
                 config=config,
                 wait_timeout=wait_timeout,
                 name=name,
@@ -308,6 +475,37 @@ class RemoteTable(Table):
             )
         )
 
+    def _is_legacy_create_index_call(
+        self,
+        first_arg: str,
+        config: Optional[IndexConfigType],
+        num_partitions: Optional[int],
+        num_sub_vectors: Optional[int],
+        vector_column_name: str,
+        accelerator: Optional[str],
+        index_cache_size: Optional[int],
+        replace: Optional[bool],
+    ) -> bool:
+        """Detect if this is a legacy create_index call."""
+        if config is not None:
+            return False
+        if any(
+            x is not None
+            for x in (
+                num_partitions,
+                num_sub_vectors,
+                accelerator,
+                index_cache_size,
+                replace,
+            )
+        ):
+            return True
+        if vector_column_name != VECTOR_COLUMN_NAME:
+            return True
+        if first_arg.lower() in KNOWN_METRICS:
+            return True
+        return False
+
     def add(
         self,
         data: DATA,
@@ -653,6 +851,11 @@ class RemoteTable(Table):
     ) -> AlterColumnsResult:
         return LOOP.run(self._table.alter_columns(*alterations))
 
+    def update_field_metadata(
+        self, *updates: dict[str, Any]
+    ) -> UpdateFieldMetadataResult:
+        return LOOP.run(self._table.update_field_metadata(*updates))
+
     def drop_columns(self, columns: Iterable[str]) -> DropColumnsResult:
         return LOOP.run(self._table.drop_columns(columns))
 
@@ -668,6 +871,10 @@ class RemoteTable(Table):
         """Not supported on LanceDB Cloud."""
         return LOOP.run(self._table.unset_lsm_write_spec())
 
+    def close_lsm_writers(self) -> None:
+        """No-op on LanceDB Cloud (no local shard writers)."""
+        return LOOP.run(self._table.close_lsm_writers())
+
     def drop_index(self, index_name: str):
         return LOOP.run(self._table.drop_index(index_name))
 

python/python/lancedb/rerankers/linear_combination.py

@@ -102,8 +102,15 @@ class LinearCombinationReranker(Reranker):
 
         combined_list = []
         for row_id, result in results.items():
+            # Convert vector distance to a relevance score in [0, 1] where
+            # higher is better.  Missing vector entries are penalised with
+            # `_invert_score(fill)` = 1 - fill (= 0.0 for the default fill=1).
             vector_score = self._invert_score(result.get("_distance", fill))
-            fts_score = result.get("_score", fill)
+            # FTS scores (BM25) are already in a "higher = more relevant" space.
+            # Missing FTS entries are penalised symmetrically: we use
+            # `1 - fill` so that the same `fill` value drives both missing-vector
+            # and missing-FTS penalties in the same direction.
+            fts_score = result.get("_score", 1 - fill)
             result["_relevance_score"] = self._combine_score(vector_score, fts_score)
             combined_list.append(result)
 
@@ -123,8 +130,12 @@ class LinearCombinationReranker(Reranker):
         return tbl
 
     def _combine_score(self, vector_score, fts_score):
-        # these scores represent distance
-        return 1 - (self.weight * vector_score + (1 - self.weight) * fts_score)
+        # Both vector_score (inverted distance) and fts_score are in a
+        # "higher = more relevant" space.  A straight weighted average gives
+        # higher _relevance_score to better matches, as expected.
+        # Previously this returned `1 - (...)` which inverted the final
+        # ranking so that the *least* relevant document ranked first.
+        return self.weight * vector_score + (1 - self.weight) * fts_score
 
     def _invert_score(self, dist: float):
         # Invert the score between relevance and distance

python/python/lancedb/table.py

@@ -154,6 +154,7 @@ if TYPE_CHECKING:
         AddColumnsResult,
         AddResult,
         AlterColumnsResult,
+        UpdateFieldMetadataResult,
         DeleteResult,
         DropColumnsResult,
         LsmWriteSpec,
@@ -174,6 +175,24 @@ if TYPE_CHECKING:
         DistanceType,
     )
 
+# Type alias for index configuration objects
+IndexConfigType = Union[
+    IvfFlat,
+    IvfPq,
+    IvfSq,
+    IvfRq,
+    HnswFlat,
+    HnswPq,
+    HnswSq,
+    BTree,
+    Bitmap,
+    LabelList,
+    FTS,
+]
+
+# Known distance metrics for legacy API detection
+KNOWN_METRICS = {"l2", "cosine", "dot", "hamming"}
+
 
 def _into_pyarrow_reader(
     data, schema: Optional[pa.Schema] = None
@@ -807,11 +826,49 @@ class Table(ABC):
         """
         raise NotImplementedError
 
+    # New unified API overload
+    @overload
     def create_index(
         self,
-        metric="l2",
-        num_partitions=256,
-        num_sub_vectors=96,
+        column: str,
+        /,
+        *,
+        config: IndexConfigType,
+        replace: bool = ...,
+        wait_timeout: Optional[timedelta] = ...,
+        name: Optional[str] = ...,
+        train: bool = ...,
+    ) -> None: ...
+
+    # Legacy API overload (deprecated)
+    @overload
+    def create_index(
+        self,
+        metric: Literal["l2", "cosine", "dot", "hamming"] = ...,
+        num_partitions: Optional[int] = ...,
+        num_sub_vectors: Optional[int] = ...,
+        vector_column_name: str = ...,
+        replace: bool = ...,
+        accelerator: Optional[str] = ...,
+        index_cache_size: Optional[int] = ...,
+        *,
+        index_type: VectorIndexType = ...,
+        wait_timeout: Optional[timedelta] = ...,
+        num_bits: int = ...,
+        max_iterations: int = ...,
+        sample_rate: int = ...,
+        m: int = ...,
+        ef_construction: int = ...,
+        name: Optional[str] = ...,
+        train: bool = ...,
+        target_partition_size: Optional[int] = ...,
+    ) -> None: ...
+
+    def create_index(
+        self,
+        metric: DistanceType = "l2",
+        num_partitions: Optional[int] = None,
+        num_sub_vectors: Optional[int] = None,
         vector_column_name: str = VECTOR_COLUMN_NAME,
         replace: bool = True,
         accelerator: Optional[str] = None,
@@ -824,46 +881,53 @@ class Table(ABC):
         sample_rate: int = 256,
         m: int = 20,
         ef_construction: int = 300,
+        config: Optional[IndexConfigType] = None,
         name: Optional[str] = None,
         train: bool = True,
         target_partition_size: Optional[int] = None,
     ):
-        """Create an index on the table.
+        """Create an index on a column.
+
+        This method supports both the new unified API and the legacy API
+        for backwards compatibility. The new API takes the column name as the
+        first positional argument and an index configuration object via
+        ``config``; the legacy API takes the distance metric as the first
+        argument plus separate ``vector_column_name`` / ``num_partitions`` /
+        etc. parameters, and emits a ``DeprecationWarning``.
 
         Parameters
         ----------
-        metric: str, default "l2"
-            The distance metric to use when creating the index.
-            Valid values are "l2", "cosine", "dot", or "hamming".
-            l2 is euclidean distance.
-            Hamming is available only for binary vectors.
-        num_partitions: int, default 256
-            The number of IVF partitions to use when creating the index.
-            Default is 256.
-        num_sub_vectors: int, default 96
-            The number of PQ sub-vectors to use when creating the index.
-            Default is 96.
-        vector_column_name: str, default "vector"
-            The vector column name to create the index.
-        replace: bool, default True
-            - If True, replace the existing index if it exists.
+        metric : str
+            For new API: the column name to index.
+            For legacy API: the distance metric ("l2", "cosine", "dot", "hamming").
+        config : IndexConfigType, optional
+            The index configuration object. If provided, uses the new unified API.
+            Can be one of: IvfFlat, IvfPq, IvfSq, IvfRq, HnswPq, HnswSq,
+            BTree, Bitmap, LabelList, FTS.
+        replace : bool, default True
+            Whether to replace an existing index on this column.
+        wait_timeout : timedelta, optional
+            Timeout to wait for async indexing to complete.
+        name : str, optional
+            Custom name for the index.
+        train : bool, default True
+            Whether to train the index with existing data.
 
-            - If False, raise an error if duplicate index exists.
-        accelerator: str, default None
-            If set, use the given accelerator to create the index.
-            Only support "cuda" for now.
-        index_cache_size : int, optional
-            The size of the index cache in number of entries. Default value is 256.
-        num_bits: int
-            The number of bits to encode sub-vectors. Only used with the IVF_PQ index.
-            Only 4 and 8 are supported.
-        wait_timeout: timedelta, optional
-            The timeout to wait if indexing is asynchronous.
-        name: str, optional
-            The name of the index. If not provided, a default name will be generated.
-        train: bool, default True
-            Whether to train the index with existing data. Vector indices always train
-            with existing data.
+        Examples
+        --------
+        New API (recommended):
+
+        >>> table.create_index(  # doctest: +SKIP
+        ...     "vector", config=IvfPq(distance_type="l2")
+        ... )
+        >>> table.create_index("category", config=BTree())  # doctest: +SKIP
+        >>> table.create_index("content", config=FTS())  # doctest: +SKIP
+
+        Legacy API (deprecated):
+
+        >>> table.create_index(  # doctest: +SKIP
+        ...     "l2", vector_column_name="vector"
+        ... )
         """
         raise NotImplementedError
 
@@ -1188,7 +1252,7 @@ class Table(ABC):
         ...      .when_not_matched_insert_all() \\
         ...      .execute(new_data)
         >>> res
-        MergeResult(version=2, num_updated_rows=2, num_inserted_rows=1, num_deleted_rows=0, num_attempts=1)
+        MergeResult(version=2, num_updated_rows=2, num_inserted_rows=1, num_deleted_rows=0, num_attempts=1, num_rows=3)
         >>> # The order of new rows is non-deterministic since we use
         >>> # a hash-join as part of this operation and so we sort here
         >>> table.to_arrow().sort_by("a").to_pandas()
@@ -1736,6 +1800,29 @@ class Table(ABC):
             version: the new version number of the table after the alteration.
         """
 
+    @abstractmethod
+    def update_field_metadata(
+        self, *updates: dict[str, Any]
+    ) -> UpdateFieldMetadataResult:
+        """
+        Update per-field (column) metadata.
+
+        Parameters
+        ----------
+        updates : dict
+            One or more dicts, each with:
+            - "path": str — dot-path to the field (e.g. "embedding" or "a.b.c").
+            - "metadata": dict[str, str | None] — keys to set; a value of ``None``
+              deletes that key.
+            - "replace": bool, optional — replace the field's whole metadata map
+              instead of merging (default False).
+
+        Returns
+        -------
+        UpdateFieldMetadataResult
+            version: the new table version after the update.
+        """
+
     @abstractmethod
     def drop_columns(self, columns: Iterable[str]) -> DropColumnsResult:
         """
@@ -2178,7 +2265,7 @@ class LanceTable(Table):
         return LOOP.run(self._table.count_rows(filter))
 
     def __repr__(self) -> str:
-        val = f"{self.__class__.__name__}(name={self.name!r}, version={self.version}"
+        val = f"{self.__class__.__name__}(name={self.name!r}"
         if self._conn.read_consistency_interval is not None:
             val += ", read_consistency_interval={!r}".format(
                 self._conn.read_consistency_interval
@@ -2250,11 +2337,51 @@ class LanceTable(Table):
             dataset, allow_pyarrow_filter=False, batch_size=batch_size
         )
 
+    # New unified API overload
+    @overload
     def create_index(
         self,
-        metric: DistanceType = "l2",
-        num_partitions=None,
-        num_sub_vectors=None,
+        column: str,
+        /,
+        *,
+        config: IndexConfigType,
+        replace: bool = ...,
+        wait_timeout: Optional[timedelta] = ...,
+        name: Optional[str] = ...,
+        train: bool = ...,
+    ) -> None: ...
+
+    # Legacy API overload (deprecated)
+    @overload
+    def create_index(
+        self,
+        metric: Literal["l2", "cosine", "dot", "hamming"] = ...,
+        num_partitions: Optional[int] = ...,
+        num_sub_vectors: Optional[int] = ...,
+        vector_column_name: str = ...,
+        replace: bool = ...,
+        accelerator: Optional[str] = ...,
+        index_cache_size: Optional[int] = ...,
+        num_bits: int = ...,
+        index_type: Literal[
+            "IVF_FLAT", "IVF_SQ", "IVF_PQ", "IVF_RQ", "IVF_HNSW_SQ", "IVF_HNSW_PQ"
+        ] = ...,
+        max_iterations: int = ...,
+        sample_rate: int = ...,
+        m: int = ...,
+        ef_construction: int = ...,
+        *,
+        wait_timeout: Optional[timedelta] = ...,
+        name: Optional[str] = ...,
+        train: bool = ...,
+        target_partition_size: Optional[int] = ...,
+    ) -> None: ...
+
+    def create_index(
+        self,
+        metric: str = "l2",
+        num_partitions: Optional[int] = None,
+        num_sub_vectors: Optional[int] = None,
         vector_column_name: str = VECTOR_COLUMN_NAME,
         replace: bool = True,
         accelerator: Optional[str] = None,
@@ -2274,47 +2401,232 @@ class LanceTable(Table):
         m: int = 20,
         ef_construction: int = 300,
         *,
+        config: Optional[IndexConfigType] = None,
+        wait_timeout: Optional[timedelta] = None,
         name: Optional[str] = None,
         train: bool = True,
         target_partition_size: Optional[int] = None,
     ):
-        """Create an index on the table."""
-        if accelerator is not None:
-            # accelerator is only supported through pylance.
-            self.to_lance().create_index(
-                column=vector_column_name,
-                index_type=index_type,
+        """Create an index on a column.
+
+        This method supports both the new unified API and the legacy API
+        for backwards compatibility. The new API takes the column name as the
+        first positional argument and an index configuration object via
+        ``config``; the legacy API takes the distance metric as the first
+        argument plus separate ``vector_column_name`` / ``num_partitions`` /
+        etc. parameters, and emits a ``DeprecationWarning``.
+
+        Parameters
+        ----------
+        metric : str
+            For new API: the column name to index.
+            For legacy API: the distance metric ("l2", "cosine", "dot", "hamming").
+        config : IndexConfigType, optional
+            The index configuration object. If provided, uses the new unified API.
+            Can be one of: IvfFlat, IvfPq, IvfSq, IvfRq, HnswPq, HnswSq,
+            BTree, Bitmap, LabelList, FTS.
+        replace : bool, default True
+            Whether to replace an existing index on this column.
+        wait_timeout : timedelta, optional
+            Timeout to wait for async indexing to complete.
+        name : str, optional
+            Custom name for the index.
+        train : bool, default True
+            Whether to train the index with existing data.
+
+        Examples
+        --------
+        New API (recommended):
+
+        >>> table.create_index(  # doctest: +SKIP
+        ...     "vector", config=IvfPq(distance_type="l2")
+        ... )
+        >>> table.create_index("category", config=BTree())  # doctest: +SKIP
+        >>> table.create_index("content", config=FTS())  # doctest: +SKIP
+
+        Legacy API (deprecated):
+
+        >>> table.create_index(  # doctest: +SKIP
+        ...     "l2", vector_column_name="vector"
+        ... )
+        """
+        # Detect whether this is a legacy API call
+        is_legacy = self._is_legacy_create_index_call(
+            metric,
+            config,
+            num_partitions,
+            num_sub_vectors,
+            vector_column_name,
+            accelerator,
+            index_cache_size,
+        )
+
+        if is_legacy:
+            warnings.warn(
+                "The create_index() API with metric/num_partitions parameters is "
+                "deprecated and will be removed in a future version. "
+                "Please migrate to the new unified API:\n"
+                "  # Old (deprecated):\n"
+                "  table.create_index('l2', vector_column_name='my_vector')\n"
+                "  # New (recommended):\n"
+                "  table.create_index('my_vector', config=IvfPq(distance_type='l2'))",
+                DeprecationWarning,
+                stacklevel=2,
+            )
+
+            # Legacy API: first arg is the distance metric
+            column = vector_column_name
+
+            # Build config from legacy parameters
+            config = self._build_vector_config_from_legacy_params(
                 metric=metric,
+                index_type=index_type,
                 num_partitions=num_partitions,
                 num_sub_vectors=num_sub_vectors,
-                replace=replace,
-                accelerator=accelerator,
-                index_cache_size=index_cache_size,
                 num_bits=num_bits,
+                max_iterations=max_iterations,
+                sample_rate=sample_rate,
                 m=m,
                 ef_construction=ef_construction,
                 target_partition_size=target_partition_size,
+                accelerator=accelerator,
             )
-            self.checkout_latest()
-            return
-        elif index_type == "IVF_FLAT":
-            config = IvfFlat(
+
+            # Handle accelerator through pylance
+            if accelerator is not None:
+                self.to_lance().create_index(
+                    column=column,
+                    index_type=index_type,
+                    metric=metric,
+                    num_partitions=num_partitions,
+                    num_sub_vectors=num_sub_vectors,
+                    replace=replace,
+                    accelerator=accelerator,
+                    index_cache_size=index_cache_size,
+                    num_bits=num_bits,
+                    m=m,
+                    ef_construction=ef_construction,
+                    target_partition_size=target_partition_size,
+                )
+                self.checkout_latest()
+                return
+        else:
+            # New API: metric is the column name
+            column = metric
+
+            # Check if config has accelerator set and dispatch to pylance
+            if config is not None and hasattr(config, "accelerator"):
+                acc = getattr(config, "accelerator", None)
+                if acc is not None:
+                    # Dispatch to pylance for GPU acceleration
+                    index_type_map = {
+                        "IvfFlat": "IVF_FLAT",
+                        "IvfSq": "IVF_SQ",
+                        "IvfPq": "IVF_PQ",
+                        "IvfRq": "IVF_RQ",
+                        "HnswPq": "IVF_HNSW_PQ",
+                        "HnswSq": "IVF_HNSW_SQ",
+                    }
+                    cfg_type = type(config).__name__
+                    lance_index_type = index_type_map.get(cfg_type, "IVF_PQ")
+
+                    self.to_lance().create_index(
+                        column=column,
+                        index_type=lance_index_type,
+                        metric=getattr(config, "distance_type", "l2"),
+                        num_partitions=getattr(config, "num_partitions", None),
+                        num_sub_vectors=getattr(config, "num_sub_vectors", None),
+                        replace=replace,
+                        accelerator=acc,
+                        num_bits=getattr(config, "num_bits", 8),
+                        m=getattr(config, "m", 20),
+                        ef_construction=getattr(config, "ef_construction", 300),
+                        target_partition_size=getattr(
+                            config, "target_partition_size", None
+                        ),
+                    )
+                    self.checkout_latest()
+                    return
+
+        return LOOP.run(
+            self._table.create_index(
+                column,
+                replace=replace,
+                config=config,
+                wait_timeout=wait_timeout,
+                name=name,
+                train=train,
+            )
+        )
+
+    def _is_legacy_create_index_call(
+        self,
+        first_arg: str,
+        config: Optional[IndexConfigType],
+        num_partitions: Optional[int],
+        num_sub_vectors: Optional[int],
+        vector_column_name: str,
+        accelerator: Optional[str],
+        index_cache_size: Optional[int],
+    ) -> bool:
+        """Detect if this is a legacy create_index call."""
+        # If config is provided, it's definitely the new API
+        if config is not None:
+            return False
+
+        # If old-style parameters were explicitly set, it's legacy
+        if any(
+            x is not None
+            for x in (num_partitions, num_sub_vectors, accelerator, index_cache_size)
+        ):
+            return True
+
+        # If vector_column_name differs from default, it's legacy
+        if vector_column_name != VECTOR_COLUMN_NAME:
+            return True
+
+        # If first arg is a known metric, assume legacy
+        if first_arg.lower() in KNOWN_METRICS:
+            return True
+
+        # Otherwise assume new API
+        return False
+
+    def _build_vector_config_from_legacy_params(
+        self,
+        metric: str,
+        index_type: str,
+        num_partitions: Optional[int],
+        num_sub_vectors: Optional[int],
+        num_bits: int,
+        max_iterations: int,
+        sample_rate: int,
+        m: int,
+        ef_construction: int,
+        target_partition_size: Optional[int],
+        accelerator: Optional[str],
+    ) -> IndexConfigType:
+        """Build an index config object from legacy parameters."""
+        if index_type == "IVF_FLAT":
+            return IvfFlat(
                 distance_type=metric,
                 num_partitions=num_partitions,
                 max_iterations=max_iterations,
                 sample_rate=sample_rate,
                 target_partition_size=target_partition_size,
+                accelerator=accelerator,
             )
         elif index_type == "IVF_SQ":
-            config = IvfSq(
+            return IvfSq(
                 distance_type=metric,
                 num_partitions=num_partitions,
                 max_iterations=max_iterations,
                 sample_rate=sample_rate,
                 target_partition_size=target_partition_size,
+                accelerator=accelerator,
             )
         elif index_type == "IVF_PQ":
-            config = IvfPq(
+            return IvfPq(
                 distance_type=metric,
                 num_partitions=num_partitions,
                 num_sub_vectors=num_sub_vectors,
@@ -2322,18 +2634,20 @@ class LanceTable(Table):
                 max_iterations=max_iterations,
                 sample_rate=sample_rate,
                 target_partition_size=target_partition_size,
+                accelerator=accelerator,
             )
         elif index_type == "IVF_RQ":
-            config = IvfRq(
+            return IvfRq(
                 distance_type=metric,
                 num_partitions=num_partitions,
                 num_bits=num_bits,
                 max_iterations=max_iterations,
                 sample_rate=sample_rate,
                 target_partition_size=target_partition_size,
+                accelerator=accelerator,
             )
         elif index_type == "IVF_HNSW_PQ":
-            config = HnswPq(
+            return HnswPq(
                 distance_type=metric,
                 num_partitions=num_partitions,
                 num_sub_vectors=num_sub_vectors,
@@ -2343,9 +2657,10 @@ class LanceTable(Table):
                 m=m,
                 ef_construction=ef_construction,
                 target_partition_size=target_partition_size,
+                accelerator=accelerator,
             )
         elif index_type == "IVF_HNSW_SQ":
-            config = HnswSq(
+            return HnswSq(
                 distance_type=metric,
                 num_partitions=num_partitions,
                 max_iterations=max_iterations,
@@ -2353,9 +2668,10 @@ class LanceTable(Table):
                 m=m,
                 ef_construction=ef_construction,
                 target_partition_size=target_partition_size,
+                accelerator=accelerator,
             )
         elif index_type == "IVF_HNSW_FLAT":
-            config = HnswFlat(
+            return HnswFlat(
                 distance_type=metric,
                 num_partitions=num_partitions,
                 max_iterations=max_iterations,
@@ -2367,16 +2683,6 @@ class LanceTable(Table):
         else:
             raise ValueError(f"Unknown index type {index_type}")
 
-        return LOOP.run(
-            self._table.create_index(
-                vector_column_name,
-                replace=replace,
-                config=config,
-                name=name,
-                train=train,
-            )
-        )
-
     def drop_index(self, name: str) -> None:
         """
         Drops an index from the table
@@ -2476,6 +2782,11 @@ class LanceTable(Table):
         """
         return LOOP.run(self._table.latest_storage_options())
 
+    @deprecation.deprecated(
+        deprecated_in="0.25.0",
+        current_version=__version__,
+        details="Use create_index() with config=BTree()/Bitmap()/LabelList() instead.",
+    )
     def create_scalar_index(
         self,
         column: str,
@@ -2484,6 +2795,12 @@ class LanceTable(Table):
         index_type: ScalarIndexType = "BTREE",
         name: Optional[str] = None,
     ):
+        """Create a scalar index on a column.
+
+        .. deprecated:: 0.25.0
+            Use :meth:`create_index` with a BTree, Bitmap, or LabelList config instead.
+            Example: ``table.create_index("column", config=BTree())``
+        """
         if index_type == "BTREE":
             config = BTree()
         elif index_type == "BITMAP":
@@ -2496,6 +2813,11 @@ class LanceTable(Table):
             self._table.create_index(column, replace=replace, config=config, name=name)
         )
 
+    @deprecation.deprecated(
+        deprecated_in="0.25.0",
+        current_version=__version__,
+        details="Use create_index() with config=FTS() instead.",
+    )
     def create_fts_index(
         self,
         field_names: Union[str, List[str]],
@@ -2519,6 +2841,12 @@ class LanceTable(Table):
         prefix_only: bool = False,
         name: Optional[str] = None,
     ):
+        """Create a full-text search index on a column.
+
+        .. deprecated:: 0.25.0
+            Use :meth:`create_index` with an FTS config instead.
+            Example: ``table.create_index("text_column", config=FTS())``
+        """
         self._ensure_no_legacy_fts_index()
 
         if use_tantivy:
@@ -3279,6 +3607,11 @@ class LanceTable(Table):
     ) -> AlterColumnsResult:
         return LOOP.run(self._table.alter_columns(*alterations))
 
+    def update_field_metadata(
+        self, *updates: dict[str, Any]
+    ) -> UpdateFieldMetadataResult:
+        return LOOP.run(self._table.update_field_metadata(*updates))
+
     def drop_columns(self, columns: Iterable[str]) -> DropColumnsResult:
         return LOOP.run(self._table.drop_columns(columns))
 
@@ -3297,6 +3630,11 @@ class LanceTable(Table):
         [`AsyncTable.unset_lsm_write_spec`][lancedb.AsyncTable.unset_lsm_write_spec]."""
         return LOOP.run(self._table.unset_lsm_write_spec())
 
+    def close_lsm_writers(self) -> None:
+        """Close cached MemWAL shard writers. See
+        [`AsyncTable.close_lsm_writers`][lancedb.AsyncTable.close_lsm_writers]."""
+        return LOOP.run(self._table.close_lsm_writers())
+
     def uses_v2_manifest_paths(self) -> bool:
         """
         Check if the table is using the new v2 manifest paths.
@@ -3328,10 +3666,18 @@ class LanceTable(Table):
         """
         LOOP.run(self._table.migrate_v2_manifest_paths())
 
+    @deprecation.deprecated(
+        deprecated_in="0.33.1",
+        current_version=__version__,
+        details="Use update_field_metadata() instead.",
+    )
     def replace_field_metadata(self, field_name: str, new_metadata: Dict[str, str]):
         """
         Replace the metadata of a field in the schema
 
+        .. deprecated:: 0.33.1
+            Use :func:`update_field_metadata` instead.
+
         Parameters
         ----------
         field_name: str
@@ -3905,6 +4251,16 @@ class AsyncTable:
         """
         await self._inner.unset_lsm_write_spec()
 
+    async def close_lsm_writers(self) -> None:
+        """Drain and close any cached MemWAL shard writers for this table.
+
+        When an LSM write spec is installed, `merge_insert` opens MemWAL shard
+        writers and caches them for reuse across calls. This closes them,
+        flushing pending data; writers reopen lazily on the next
+        `merge_insert`. It is a no-op when no writers are cached.
+        """
+        await self._inner.close_lsm_writers()
+
     @property
     def name(self) -> str:
         """The name of the table."""
@@ -4355,7 +4711,7 @@ class AsyncTable:
         ...      .when_not_matched_insert_all() \\
         ...      .execute(new_data)
         >>> res
-        MergeResult(version=2, num_updated_rows=2, num_inserted_rows=1, num_deleted_rows=0, num_attempts=1)
+        MergeResult(version=2, num_updated_rows=2, num_inserted_rows=1, num_deleted_rows=0, num_attempts=1, num_rows=3)
         >>> # The order of new rows is non-deterministic since we use
         >>> # a hash-join as part of this operation and so we sort here
         >>> table.to_arrow().sort_by("a").to_pandas()
@@ -4735,6 +5091,8 @@ class AsyncTable:
                 when_not_matched_by_source_condition=merge._when_not_matched_by_source_condition,
                 timeout=merge._timeout,
                 use_index=merge._use_index,
+                use_lsm_write=merge._use_lsm_write,
+                validate_single_shard=merge._validate_single_shard,
             ),
         )
 
@@ -4913,6 +5271,13 @@ class AsyncTable:
         """
         return await self._inner.alter_columns(alterations)
 
+    async def update_field_metadata(
+        self, *updates: dict[str, Any]
+    ) -> UpdateFieldMetadataResult:
+        """Update per-field metadata. See
+        [`Table.update_field_metadata`][lancedb.table.Table.update_field_metadata]."""
+        return await self._inner.update_field_metadata(updates)
+
     async def drop_columns(self, columns: Iterable[str]):
         """
         Drop columns from the table.
@@ -5197,12 +5562,20 @@ class AsyncTable:
         """
         await self._inner.migrate_manifest_paths_v2()
 
+    @deprecation.deprecated(
+        deprecated_in="0.33.1",
+        current_version=__version__,
+        details="Use update_field_metadata() instead.",
+    )
     async def replace_field_metadata(
         self, field_name: str, new_metadata: dict[str, str]
     ):
         """
         Replace the metadata of a field in the schema
 
+        .. deprecated:: 0.33.1
+            Use :func:`update_field_metadata` instead.
+
         Parameters
         ----------
         field_name: str

python/python/lancedb/util.py

@@ -10,7 +10,7 @@ import pathlib
 import warnings
 from datetime import date, datetime
 from functools import singledispatch
-from typing import Tuple, Union, Optional, Any
+from typing import Tuple, Union, Optional, Any, List
 from urllib.parse import urlparse
 
 import numpy as np
@@ -189,7 +189,33 @@ def flatten_columns(tbl: pa.Table, flatten: Optional[Union[int, bool]] = None):
     return tbl
 
 
-def inf_vector_column_query(schema: pa.Schema) -> str:
+def _format_field_path(path: List[str]) -> str:
+    def format_segment(segment: str) -> str:
+        if all(char.isalnum() or char == "_" for char in segment):
+            return segment
+        return f"`{segment.replace('`', '``')}`"
+
+    return ".".join(format_segment(segment) for segment in path)
+
+
+def _iter_vector_columns(
+    field: pa.Field, path: List[str], dim: Optional[int] = None
+) -> List[str]:
+    field_path = [*path, field.name]
+    if is_vector_column(field.type):
+        vector_dim = infer_vector_column_dim(field.type)
+        if dim is None or vector_dim == dim:
+            return [_format_field_path(field_path)]
+        return []
+    if pa.types.is_struct(field.type):
+        columns = []
+        for idx in range(field.type.num_fields):
+            columns.extend(_iter_vector_columns(field.type.field(idx), field_path, dim))
+        return columns
+    return []
+
+
+def inf_vector_column_query(schema: pa.Schema, dim: Optional[int] = None) -> str:
     """
     Get the vector column name
 
@@ -202,26 +228,21 @@ def inf_vector_column_query(schema: pa.Schema) -> str:
     -------
     str: the vector column name.
     """
-    vector_col_name = ""
-    vector_col_count = 0
-    for field_name in schema.names:
-        field = schema.field(field_name)
-        if is_vector_column(field.type):
-            vector_col_count += 1
-            if vector_col_count > 1:
-                raise ValueError(
-                    "Schema has more than one vector column. "
-                    "Please specify the vector column name "
-                    "for vector search"
-                )
-            elif vector_col_count == 1:
-                vector_col_name = field_name
-    if vector_col_count == 0:
+    vector_col_names = []
+    for field in schema:
+        vector_col_names.extend(_iter_vector_columns(field, [], dim))
+    if len(vector_col_names) > 1:
+        raise ValueError(
+            "Schema has more than one vector column. "
+            "Please specify the vector column name "
+            f"for vector search. Candidates: {vector_col_names}"
+        )
+    if len(vector_col_names) == 0:
         raise ValueError(
             "There is no vector column in the data. "
             "Please specify the vector column name for vector search"
         )
-    return vector_col_name
+    return vector_col_names[0]
 
 
 def is_vector_column(data_type: pa.DataType) -> bool:
@@ -247,6 +268,29 @@ def is_vector_column(data_type: pa.DataType) -> bool:
     return False
 
 
+def infer_vector_column_dim(data_type: pa.DataType) -> Optional[int]:
+    if pa.types.is_fixed_size_list(data_type):
+        return data_type.list_size
+    if pa.types.is_list(data_type):
+        return infer_vector_column_dim(data_type.value_type)
+    return None
+
+
+def _query_vector_dim(query: Optional[Any]) -> Optional[int]:
+    if query is None:
+        return None
+    if isinstance(query, np.ndarray):
+        if query.ndim == 0:
+            return None
+        return query.shape[-1]
+    if isinstance(query, list) and query:
+        first = query[0]
+        if isinstance(first, (list, tuple, np.ndarray)):
+            return len(first)
+        return len(query)
+    return None
+
+
 def infer_vector_column_name(
     schema: pa.Schema,
     query_type: str,
@@ -262,7 +306,9 @@ def infer_vector_column_name(
 
     if query is not None or query_type == "hybrid":
         try:
-            vector_column_name = inf_vector_column_query(schema)
+            vector_column_name = inf_vector_column_query(
+                schema, dim=_query_vector_dim(query)
+            )
         except Exception as e:
             raise e
 

python/python/tests/docs/test_merge_insert.py

@@ -57,7 +57,7 @@ async def test_upsert_async(mem_db_async):
     await table.count_rows()  # 3
     res
     # MergeResult(version=2, num_updated_rows=1,
-    # num_inserted_rows=1, num_deleted_rows=0)
+    # num_inserted_rows=1, num_deleted_rows=0, num_rows=2)
     # --8<-- [end:upsert_basic_async]
     assert await table.count_rows() == 3
     assert res.version == 2
@@ -86,7 +86,7 @@ def test_insert_if_not_exists(mem_db):
     table.count_rows()  # 3
     res
     # MergeResult(version=2, num_updated_rows=0,
-    # num_inserted_rows=1, num_deleted_rows=0)
+    # num_inserted_rows=1, num_deleted_rows=0, num_rows=1)
     # --8<-- [end:insert_if_not_exists]
     assert table.count_rows() == 3
     assert res.version == 2
@@ -116,7 +116,7 @@ async def test_insert_if_not_exists_async(mem_db_async):
     await table.count_rows()  # 3
     res
     # MergeResult(version=2, num_updated_rows=0,
-    # num_inserted_rows=1, num_deleted_rows=0)
+    # num_inserted_rows=1, num_deleted_rows=0, num_rows=1)
     # --8<-- [end:insert_if_not_exists]
     assert await table.count_rows() == 3
     assert res.version == 2
@@ -150,7 +150,7 @@ def test_replace_range(mem_db):
     table.count_rows("doc_id = 1")  # 1
     res
     # MergeResult(version=2, num_updated_rows=1,
-    # num_inserted_rows=0, num_deleted_rows=1)
+    # num_inserted_rows=0, num_deleted_rows=1, num_rows=1)
     # --8<-- [end:insert_if_not_exists]
     assert table.count_rows("doc_id = 1") == 1
     assert res.version == 2
@@ -185,7 +185,7 @@ async def test_replace_range_async(mem_db_async):
     await table.count_rows("doc_id = 1")  # 1
     res
     # MergeResult(version=2, num_updated_rows=1,
-    # num_inserted_rows=0, num_deleted_rows=1)
+    # num_inserted_rows=0, num_deleted_rows=1, num_rows=1)
     # --8<-- [end:insert_if_not_exists]
     assert await table.count_rows("doc_id = 1") == 1
     assert res.version == 2

python/python/tests/test_db.py

@@ -6,6 +6,7 @@ import re
 import sys
 from datetime import timedelta
 import os
+from types import SimpleNamespace
 
 import lancedb
 import numpy as np
@@ -188,6 +189,43 @@ def test_table_names(tmp_db: lancedb.DBConnection):
     assert len(result) == 3
 
 
+def test_db_contains_and_len_include_all_table_name_pages(tmp_db: lancedb.DBConnection):
+    for idx in range(20):
+        tmp_db.create_table(f"table_{idx}", data=[{"id": idx}])
+
+    assert len(tmp_db) == 20
+    for idx in range(20):
+        assert f"table_{idx}" in tmp_db
+    assert "does_not_exist" not in tmp_db
+
+
+def test_db_contains_stops_after_matching_table_page(
+    tmp_db: lancedb.DBConnection, monkeypatch
+):
+    calls = []
+    pages = {
+        None: SimpleNamespace(tables=["table_0", "table_1"], page_token="next"),
+        "next": SimpleNamespace(tables=["table_2"], page_token=None),
+    }
+
+    def list_tables(*, page_token=None, **_kwargs):
+        calls.append(page_token)
+        return pages[page_token]
+
+    monkeypatch.setattr(tmp_db, "list_tables", list_tables)
+
+    assert "table_1" in tmp_db
+    assert calls == [None]
+
+    calls.clear()
+    assert "table_2" in tmp_db
+    assert calls == [None, "next"]
+
+    calls.clear()
+    assert len(tmp_db) == 3
+    assert calls == [None, "next"]
+
+
 @pytest.mark.asyncio
 async def test_table_names_async(tmp_path):
     db = lancedb.connect(tmp_path)
@@ -428,7 +466,8 @@ async def test_create_table_v2_manifest_paths_async(tmp_path):
     assert await tbl.uses_v2_manifest_paths()
     manifests_dir = tmp_path / "test_v2_manifest_paths.lance" / "_versions"
     for manifest in os.listdir(manifests_dir):
-        assert re.match(r"\d{20}\.manifest", manifest)
+        if manifest.endswith(".manifest"):
+            assert re.match(r"\d{20}\.manifest", manifest)
 
     # Start a table in V1 mode then migrate
     tbl = await db_no_v2_paths.create_table(
@@ -438,13 +477,15 @@ async def test_create_table_v2_manifest_paths_async(tmp_path):
     assert not await tbl.uses_v2_manifest_paths()
     manifests_dir = tmp_path / "test_v2_migration.lance" / "_versions"
     for manifest in os.listdir(manifests_dir):
-        assert re.match(r"\d\.manifest", manifest)
+        if manifest.endswith(".manifest"):
+            assert re.match(r"\d\.manifest", manifest)
 
     await tbl.migrate_manifest_paths_v2()
     assert await tbl.uses_v2_manifest_paths()
 
     for manifest in os.listdir(manifests_dir):
-        assert re.match(r"\d{20}\.manifest", manifest)
+        if manifest.endswith(".manifest"):
+            assert re.match(r"\d{20}\.manifest", manifest)
 
 
 @pytest.mark.asyncio

python/python/tests/test_fts.py

@@ -215,11 +215,12 @@ def test_reject_legacy_tantivy_index(table):
 
 @pytest.mark.parametrize("with_position", [True, False])
 def test_create_inverted_index(table, with_position):
-    table.create_fts_index(
-        "text",
-        with_position=with_position,
-        name="custom_fts_index",
-    )
+    with pytest.warns(DeprecationWarning, match="create_fts_index"):
+        table.create_fts_index(
+            "text",
+            with_position=with_position,
+            name="custom_fts_index",
+        )
     indices = table.list_indices()
     fts_indices = [i for i in indices if i.index_type == "FTS"]
     assert any(i.name == "custom_fts_index" for i in fts_indices)
@@ -563,7 +564,7 @@ def test_create_index_multiple_columns(tmp_path, table):
 
 
 def test_nested_schema(tmp_path, table):
-    table.create_fts_index("nested.text")
+    table.create_fts_index("nested.text", with_position=True)
     indices = table.list_indices()
     assert len(indices) == 1
     assert indices[0].index_type == "FTS"
@@ -577,6 +578,98 @@ def test_nested_schema(tmp_path, table):
     assert len(results) > 0
     assert all("puppy" in row["nested"]["text"] for row in results)
 
+    results = table.search(MatchQuery("puppy", "nested.text")).limit(5).to_list()
+    assert len(results) > 0
+    assert all("puppy" in row["nested"]["text"] for row in results)
+
+    phrase_results = (
+        table.search(PhraseQuery("puppy runs", "nested.text")).limit(5).to_list()
+    )
+    assert len(phrase_results) > 0
+    assert all("puppy runs" in row["nested"]["text"] for row in phrase_results)
+
+    hybrid_results = (
+        table.search(query_type="hybrid", fts_columns="nested.text")
+        .vector([0 for _ in range(128)])
+        .text("puppy")
+        .limit(5)
+        .to_list()
+    )
+    assert len(hybrid_results) > 0
+
+
+@pytest.mark.asyncio
+async def test_nested_schema_async(async_table):
+    await async_table.create_index("nested.text", config=FTS(with_position=True))
+    indices = await async_table.list_indices()
+    assert len(indices) == 1
+    assert indices[0].index_type == "FTS"
+    assert indices[0].columns == ["nested.text"]
+
+    results = await (
+        async_table.query()
+        .nearest_to_text("puppy", columns="nested.text")
+        .limit(5)
+        .to_list()
+    )
+    assert len(results) > 0
+    assert all("puppy" in row["nested"]["text"] for row in results)
+
+    results = await (
+        async_table.query()
+        .nearest_to_text(MatchQuery("puppy", "nested.text"))
+        .limit(5)
+        .to_list()
+    )
+    assert len(results) > 0
+    assert all("puppy" in row["nested"]["text"] for row in results)
+
+    phrase_results = await (
+        async_table.query()
+        .nearest_to_text(PhraseQuery("puppy runs", "nested.text"))
+        .limit(5)
+        .to_list()
+    )
+    assert len(phrase_results) > 0
+    assert all("puppy runs" in row["nested"]["text"] for row in phrase_results)
+
+    hybrid_results = await (
+        async_table.query()
+        .nearest_to([0 for _ in range(128)])
+        .nearest_to_text("puppy", columns="nested.text")
+        .limit(5)
+        .to_list()
+    )
+    assert len(hybrid_results) > 0
+
+
+def test_nested_schema_rejects_invalid_fts_fields(tmp_path):
+    db = ldb.connect(tmp_path)
+    data = pa.table(
+        {
+            "payload": pa.array(
+                [
+                    {"text": "puppy runs", "count": 1},
+                    {"text": "car drives", "count": 2},
+                ]
+            ),
+            "vector": pa.array(
+                [[0.1, 0.1], [0.2, 0.2]],
+                type=pa.list_(pa.float32(), list_size=2),
+            ),
+        }
+    )
+    table = db.create_table("test", data=data)
+
+    with pytest.raises(ValueError, match="FTS index cannot be created.*payload"):
+        table.create_fts_index("payload")
+
+    with pytest.raises(ValueError, match="FTS index cannot be created.*count"):
+        table.create_fts_index("payload.count")
+
+    with pytest.raises(ValueError, match="Field path `payload.missing` not found"):
+        table.create_fts_index("payload.missing")
+
 
 def test_search_index_with_filter(table):
     table.create_fts_index("text")

python/python/tests/test_index.py

@@ -105,6 +105,46 @@ async def test_create_scalar_index(some_table: AsyncTable):
     assert len(indices) == 0
 
 
+@pytest.mark.asyncio
+async def test_create_nested_scalar_index_lists_canonical_paths(db_async):
+    metadata_type = pa.struct(
+        [
+            pa.field("user_id", pa.int32()),
+            pa.field("user.id", pa.int32()),
+        ]
+    )
+    data = pa.Table.from_arrays(
+        [
+            pa.array([1, 2, 3], type=pa.int32()),
+            pa.array(
+                [
+                    {"user_id": 10, "user.id": 100},
+                    {"user_id": 20, "user.id": 200},
+                    {"user_id": 30, "user.id": 300},
+                ],
+                type=metadata_type,
+            ),
+        ],
+        names=["user_id", "metadata"],
+    )
+    table = await db_async.create_table("nested_scalar_index", data)
+
+    await table.create_index("user_id", config=BTree(), name="top_user_id_idx")
+    await table.create_index(
+        "metadata.user_id", config=BTree(), name="nested_user_id_idx"
+    )
+    await table.create_index(
+        "metadata.`user.id`", config=BTree(), name="escaped_user_id_idx"
+    )
+
+    columns_by_name = {
+        index.name: index.columns for index in await table.list_indices()
+    }
+    assert columns_by_name["top_user_id_idx"] == ["user_id"]
+    assert columns_by_name["nested_user_id_idx"] == ["metadata.user_id"]
+    assert columns_by_name["escaped_user_id_idx"] == ["metadata.`user.id`"]
+
+
 @pytest.mark.asyncio
 async def test_create_fixed_size_binary_index(some_table: AsyncTable):
     await some_table.create_index("fsb", config=BTree())
@@ -122,12 +162,13 @@ async def test_create_bitmap_index(some_table: AsyncTable):
     await some_table.create_index("data", config=Bitmap())
     indices = await some_table.list_indices()
     assert len(indices) == 3
+    # list_indices returns indices in alphabetical order by name
     assert indices[0].index_type == "Bitmap"
-    assert indices[0].columns == ["id"]
+    assert indices[0].columns == ["data"]
     assert indices[1].index_type == "Bitmap"
-    assert indices[1].columns == ["is_active"]
+    assert indices[1].columns == ["id"]
     assert indices[2].index_type == "Bitmap"
-    assert indices[2].columns == ["data"]
+    assert indices[2].columns == ["is_active"]
 
     index_name = indices[0].name
     stats = await some_table.index_stats(index_name)

python/python/tests/test_lsm_write_spec.py

@@ -40,16 +40,6 @@ def _make_table(tmp_path):
 def test_set_lsm_write_spec_validates(tmp_path):
     _db, table = _make_table(tmp_path)
 
-    # No PK set yet.
-    with pytest.raises(Exception, match="primary key"):
-        table.set_lsm_write_spec(LsmWriteSpec.bucket("id", 4))
-
-    table.set_unenforced_primary_key("id")
-
-    # Column mismatch.
-    with pytest.raises(Exception, match="match"):
-        table.set_lsm_write_spec(LsmWriteSpec.bucket("v", 4))
-
     # Out-of-range num_buckets.
     with pytest.raises(Exception, match="num_buckets"):
         table.set_lsm_write_spec(LsmWriteSpec.bucket("id", 0))
@@ -70,7 +60,6 @@ def test_unset_lsm_write_spec(tmp_path):
         table.unset_lsm_write_spec()
 
     # Install a spec, then remove it; afterwards a fresh spec can be set.
-    table.set_unenforced_primary_key("id")
     table.set_lsm_write_spec(LsmWriteSpec.bucket("id", 4))
     table.unset_lsm_write_spec()
     # A second unset errors — there is no spec left to remove.

python/python/tests/test_merge_insert_lsm.py

@@ -0,0 +1,196 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright The LanceDB Authors
+
+"""Tests for the MemWAL LSM ``merge_insert`` dispatch."""
+
+from datetime import timedelta
+
+import lancedb
+import pyarrow as pa
+import pytest
+from lancedb._lancedb import LsmWriteSpec
+
+SCHEMA = pa.schema(
+    [
+        pa.field("id", pa.int64(), nullable=False),
+        pa.field("value", pa.int64(), nullable=False),
+    ]
+)
+
+REGION_SCHEMA = pa.schema(
+    [
+        pa.field("id", pa.int64(), nullable=False),
+        pa.field("region", pa.utf8(), nullable=False),
+    ]
+)
+
+
+def _reader(ids):
+    batch = pa.RecordBatch.from_arrays(
+        [
+            pa.array(ids, type=pa.int64()),
+            pa.array(list(range(len(ids))), type=pa.int64()),
+        ],
+        schema=SCHEMA,
+    )
+    return pa.RecordBatchReader.from_batches(SCHEMA, [batch])
+
+
+def _region_reader(rows):
+    batch = pa.RecordBatch.from_arrays(
+        [
+            pa.array([row[0] for row in rows], type=pa.int64()),
+            pa.array([row[1] for row in rows], type=pa.utf8()),
+        ],
+        schema=REGION_SCHEMA,
+    )
+    return pa.RecordBatchReader.from_batches(REGION_SCHEMA, [batch])
+
+
+def _bucket_table(tmp_path):
+    """A table with ``id`` as the primary key and a single-bucket LSM spec."""
+    db = lancedb.connect(tmp_path, read_consistency_interval=timedelta(seconds=0))
+    table = db.create_table("t", _reader([1, 2, 3]))
+    table.set_unenforced_primary_key("id")
+    # num_buckets = 1: every row routes to the single bucket.
+    table.set_lsm_write_spec(LsmWriteSpec.bucket("id", 1))
+    return table
+
+
+def test_lsm_merge_insert_bucket(tmp_path):
+    table = _bucket_table(tmp_path)
+    # Empty `on` defaults to the primary key.
+    result = (
+        table.merge_insert([])
+        .when_matched_update_all()
+        .when_not_matched_insert_all()
+        .execute(_reader([3, 4, 5]))
+    )
+    # LSM path: rows go to the MemWAL, so only num_rows is populated.
+    assert result.num_rows == 3
+    assert result.version == 0
+    assert result.num_inserted_rows == 0
+    assert result.num_updated_rows == 0
+
+
+def test_lsm_merge_insert_unsharded(tmp_path):
+    db = lancedb.connect(tmp_path, read_consistency_interval=timedelta(seconds=0))
+    table = db.create_table("t", _reader([1, 2, 3]))
+    table.set_unenforced_primary_key("id")
+    table.set_lsm_write_spec(LsmWriteSpec.unsharded())
+    result = (
+        table.merge_insert("id")
+        .when_matched_update_all()
+        .when_not_matched_insert_all()
+        .execute(_reader([10, 11, 12, 13]))
+    )
+    assert result.num_rows == 4
+
+
+def test_lsm_merge_insert_identity(tmp_path):
+    db = lancedb.connect(tmp_path, read_consistency_interval=timedelta(seconds=0))
+    table = db.create_table("t", _region_reader([(1, "us"), (2, "us")]))
+    table.set_unenforced_primary_key("id")
+    table.set_lsm_write_spec(LsmWriteSpec.identity("region"))
+    # All rows share one identity value, so they route to one shard.
+    result = (
+        table.merge_insert([])
+        .when_matched_update_all()
+        .when_not_matched_insert_all()
+        .execute(_region_reader([(3, "us"), (4, "us")]))
+    )
+    assert result.num_rows == 2
+
+
+def test_lsm_merge_insert_use_lsm_write_false(tmp_path):
+    table = _bucket_table(tmp_path)  # rows id = 1, 2, 3
+    # use_lsm_write(False) opts out: the standard path runs and commits.
+    result = (
+        table.merge_insert("id")
+        .when_not_matched_insert_all()
+        .use_lsm_write(False)
+        .execute(_reader([3, 4, 5]))
+    )
+    assert result.num_inserted_rows == 2
+    assert table.count_rows() == 5
+
+
+def test_lsm_merge_insert_validate_single_shard_off(tmp_path):
+    table = _bucket_table(tmp_path)
+    result = (
+        table.merge_insert([])
+        .when_matched_update_all()
+        .when_not_matched_insert_all()
+        .validate_single_shard(False)
+        .execute(_reader([6, 7, 8]))
+    )
+    assert result.num_rows == 3
+
+
+def test_lsm_merge_insert_use_lsm_write_true_requires_spec(tmp_path):
+    # A table with a primary key but no LSM write spec installed.
+    db = lancedb.connect(tmp_path, read_consistency_interval=timedelta(seconds=0))
+    table = db.create_table("t", _reader([1, 2, 3]))
+    table.set_unenforced_primary_key("id")
+    with pytest.raises(Exception, match="use_lsm_write"):
+        (
+            table.merge_insert("id")
+            .when_matched_update_all()
+            .when_not_matched_insert_all()
+            .use_lsm_write(True)
+            .execute(_reader([4]))
+        )
+
+
+def test_lsm_merge_insert_rejects_on_not_primary_key(tmp_path):
+    table = _bucket_table(tmp_path)
+    with pytest.raises(Exception, match="primary key"):
+        (
+            table.merge_insert("value")
+            .when_matched_update_all()
+            .when_not_matched_insert_all()
+            .execute(_reader([1]))
+        )
+
+
+def test_lsm_merge_insert_rejects_non_upsert(tmp_path):
+    table = _bucket_table(tmp_path)
+    # Insert-only (no when_matched_update_all) is not the upsert shape.
+    with pytest.raises(Exception, match="upsert"):
+        table.merge_insert([]).when_not_matched_insert_all().execute(_reader([4]))
+
+
+def test_lsm_close_writers(tmp_path):
+    table = _bucket_table(tmp_path)
+    (
+        table.merge_insert([])
+        .when_matched_update_all()
+        .when_not_matched_insert_all()
+        .execute(_reader([7, 8]))
+    )
+    table.close_lsm_writers()
+    # The writer reopens lazily on the next merge_insert.
+    result = (
+        table.merge_insert([])
+        .when_matched_update_all()
+        .when_not_matched_insert_all()
+        .execute(_reader([9]))
+    )
+    assert result.num_rows == 1
+
+
+@pytest.mark.asyncio
+async def test_async_lsm_merge_insert(tmp_path):
+    db = await lancedb.connect_async(
+        tmp_path, read_consistency_interval=timedelta(seconds=0)
+    )
+    table = await db.create_table("t", _reader([1, 2, 3]))
+    await table.set_unenforced_primary_key("id")
+    await table.set_lsm_write_spec(LsmWriteSpec.bucket("id", 1))
+
+    builder = (
+        table.merge_insert([]).when_matched_update_all().when_not_matched_insert_all()
+    )
+    result = await builder.execute(_reader([3, 4, 5]))
+    assert result.num_rows == 3
+    await table.close_lsm_writers()

python/python/tests/test_query.py

@@ -1512,6 +1512,37 @@ def test_take_queries(tmp_path):
     ]
 
 
+def test_take_queries_to_batches(tmp_path):
+    # Regression test for the sync take-query path: `to_batches` previously
+    # raised ``AttributeError: 'AsyncTakeQuery' object has no attribute
+    # 'execute'`` because the inherited ``BaseQueryBuilder.to_batches`` called
+    # ``execute`` on the async wrapper instead of the native query.
+    db = lancedb.connect(tmp_path)
+    data = pa.table({"idx": list(range(100)), "label": [str(i) for i in range(100)]})
+    table = db.create_table("test", data)
+
+    # Take by offset → to_batches
+    rs = list(table.take_offsets([5, 2, 17]).to_batches())
+    assert all(isinstance(b, pa.RecordBatch) for b in rs)
+    assert sum(b.num_rows for b in rs) == 3
+    assert sorted(v for b in rs for v in b.column("idx").to_pylist()) == [2, 5, 17]
+
+    # Take by row id → to_batches
+    rs = list(table.take_row_ids([5, 2, 17]).to_batches())
+    assert all(isinstance(b, pa.RecordBatch) for b in rs)
+    assert sum(b.num_rows for b in rs) == 3
+    assert sorted(v for b in rs for v in b.column("idx").to_pylist()) == [2, 5, 17]
+
+    # Take with select projection → to_batches preserves the projection
+    rs = list(table.take_row_ids([5, 2, 17]).select(["label"]).to_batches())
+    assert all(b.schema.names == ["label"] for b in rs)
+    assert sorted(v for b in rs for v in b.column("label").to_pylist()) == [
+        "17",
+        "2",
+        "5",
+    ]
+
+
 def test_getitems(tmp_path):
     db = lancedb.connect(tmp_path)
     data = pa.table(

python/python/tests/test_remote_db.py

@@ -1,12 +1,13 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright The LanceDB Authors
-import re
 from concurrent.futures import ThreadPoolExecutor
 import contextlib
 from datetime import timedelta
 import http.server
 import json
 import multiprocessing as mp
+import pickle
+import re
 import sys
 import threading
 import time
@@ -171,6 +172,155 @@ def test_table_len_sync():
         assert len(table) == 1
 
 
+def test_remote_connection_serializes():
+    def handler(request):
+        request.send_response(200)
+        request.send_header("Content-Type", "application/json")
+        request.end_headers()
+        request.wfile.write(b'{"tables": []}')
+
+    with mock_lancedb_connection(handler) as db:
+        serialized = json.loads(db.serialize())
+        assert isinstance(serialized["client_config"], dict)
+        restored = lancedb.deserialize_conn(db.serialize())
+        assert restored.table_names() == []
+
+
+def test_remote_table_is_picklable():
+    def handler(request):
+        request.close_connection = True
+        if request.path == "/v1/table/test/describe/":
+            request.send_response(200)
+            request.send_header("Content-Type", "application/json")
+            request.end_headers()
+            payload = json.dumps(
+                {
+                    "version": 1,
+                    "schema": {
+                        "fields": [
+                            {"name": "id", "type": {"type": "int64"}, "nullable": False}
+                        ]
+                    },
+                }
+            )
+            request.wfile.write(payload.encode())
+        elif request.path == "/v1/table/test/count_rows/":
+            request.send_response(200)
+            request.send_header("Content-Type", "application/json")
+            request.end_headers()
+            request.wfile.write(b"3")
+        else:
+            request.send_response(404)
+            request.end_headers()
+
+    with mock_lancedb_connection(handler) as db:
+        table = db.open_table("test")
+        restored = pickle.loads(pickle.dumps(table))
+        assert restored.count_rows() == 3
+
+
+def test_remote_table_open_does_not_require_picklable_client_config():
+    from lancedb.remote import HeaderProvider
+
+    class LocalHeaderProvider(HeaderProvider):
+        def get_headers(self):
+            return {"X-Test-Header": "present"}
+
+    def handler(request):
+        request.close_connection = True
+        assert request.headers.get("X-Test-Header") == "present"
+        if request.path == "/v1/table/test/describe/":
+            request.send_response(200)
+            request.send_header("Content-Type", "application/json")
+            request.end_headers()
+            request.wfile.write(b'{"version": 1, "schema": {"fields": []}}')
+        elif request.path == "/v1/table/test/count_rows/":
+            request.send_response(200)
+            request.send_header("Content-Type", "application/json")
+            request.end_headers()
+            request.wfile.write(b"3")
+        else:
+            request.send_response(404)
+            request.end_headers()
+
+    with http.server.HTTPServer(
+        ("localhost", 0), make_mock_http_handler(handler)
+    ) as server:
+        port = server.server_address[1]
+        handle = threading.Thread(target=server.serve_forever)
+        handle.start()
+        try:
+            db = lancedb.connect(
+                "db://dev",
+                api_key="fake",
+                host_override=f"http://localhost:{port}",
+                client_config={
+                    "retry_config": {"retries": 0},
+                    "timeout_config": {"connect_timeout": 2, "read_timeout": 2},
+                    "header_provider": LocalHeaderProvider(),
+                },
+            )
+            table = db.open_table("test")
+            assert table.count_rows() == 3
+            with pytest.raises(ValueError, match="header_provider"):
+                pickle.dumps(table)
+        finally:
+            server.shutdown()
+            handle.join()
+
+
+def test_remote_permutation_is_picklable():
+    from lancedb.permutation import Permutation
+
+    rows = list(range(10))
+
+    def handler(request):
+        request.close_connection = True
+        if request.path == "/v1/table/test/describe/":
+            request.send_response(200)
+            request.send_header("Content-Type", "application/json")
+            request.end_headers()
+            payload = json.dumps(
+                {
+                    "version": 1,
+                    "schema": {
+                        "fields": [
+                            {"name": "a", "type": {"type": "int64"}, "nullable": False}
+                        ]
+                    },
+                }
+            )
+            request.wfile.write(payload.encode())
+        elif request.path == "/v1/table/test/count_rows/":
+            request.send_response(200)
+            request.send_header("Content-Type", "application/json")
+            request.end_headers()
+            request.wfile.write(str(len(rows)).encode())
+        elif request.path == "/v1/table/test/query/":
+            content_len = int(request.headers.get("Content-Length"))
+            body = json.loads(request.rfile.read(content_len))
+            if "filter" in body:
+                match = re.search(r"_rowoffset in \((.*?)\)", body["filter"])
+                offsets = [int(offset.strip()) for offset in match.group(1).split(",")]
+            else:
+                offsets = rows
+            table = pa.table({"a": [rows[offset] for offset in offsets]})
+
+            request.send_response(200)
+            request.send_header("Content-Type", "application/vnd.apache.arrow.file")
+            request.end_headers()
+            with pa.ipc.new_file(request.wfile, schema=table.schema) as writer:
+                writer.write_table(table)
+        else:
+            request.send_response(404)
+            request.end_headers()
+
+    with mock_lancedb_connection(handler) as db:
+        permutation = Permutation.identity(db.open_table("test"))
+        restored = pickle.loads(pickle.dumps(permutation))
+        assert restored.__getitems__([0, 2, 4]) == [{"a": 0}, {"a": 2}, {"a": 4}]
+
+
 def test_create_table_exist_ok():
     def handler(request):
         if request.path == "/v1/table/test/create/?mode=exist_ok":
@@ -362,6 +512,22 @@ def test_table_create_indices():
                     schema=dict(
                         fields=[
                             dict(name="id", type={"type": "int64"}, nullable=False),
+                            dict(name="text", type={"type": "string"}, nullable=False),
+                            dict(
+                                name="vector",
+                                type={
+                                    "type": "fixed_size_list",
+                                    "fields": [
+                                        dict(
+                                            name="item",
+                                            type={"type": "float"},
+                                            nullable=True,
+                                        )
+                                    ],
+                                    "length": 2,
+                                },
+                                nullable=False,
+                            ),
                         ]
                     ),
                 )
@@ -420,22 +586,25 @@ def test_table_create_indices():
         # This is a smoke-test.
         table = db.create_table("test", [{"id": 1}])
 
-        # Test create_scalar_index with custom name
-        table.create_scalar_index(
-            "id", wait_timeout=timedelta(seconds=2), name="custom_scalar_idx"
-        )
+        # Test create_scalar_index with custom name (legacy method)
+        with pytest.warns(DeprecationWarning, match="create_scalar_index"):
+            table.create_scalar_index(
+                "id", wait_timeout=timedelta(seconds=2), name="custom_scalar_idx"
+            )
 
-        # Test create_fts_index with custom name
-        table.create_fts_index(
-            "text", wait_timeout=timedelta(seconds=2), name="custom_fts_idx"
-        )
+        # Test create_fts_index with custom name (legacy method)
+        with pytest.warns(DeprecationWarning, match="create_fts_index"):
+            table.create_fts_index(
+                "text", wait_timeout=timedelta(seconds=2), name="custom_fts_idx"
+            )
 
-        # Test create_index with custom name
-        table.create_index(
-            vector_column_name="vector",
-            wait_timeout=timedelta(seconds=10),
-            name="custom_vector_idx",
-        )
+        # Test create_index with custom name (legacy form: vector_column_name kwarg)
+        with pytest.warns(DeprecationWarning, match="create_index"):
+            table.create_index(
+                vector_column_name="vector",
+                wait_timeout=timedelta(seconds=10),
+                name="custom_vector_idx",
+            )
 
         # Validate that the name parameter was passed correctly in requests
         assert len(received_requests) == 3
@@ -464,6 +633,98 @@ def test_table_create_indices():
         table.drop_index("custom_fts_idx")
 
 
+def test_remote_create_index_new_api():
+    received_requests = []
+
+    def handler(request):
+        if request.path == "/v1/table/test/create_index/":
+            content_len = int(request.headers.get("Content-Length", 0))
+            body = request.rfile.read(content_len) if content_len > 0 else b""
+            received_requests.append(json.loads(body) if body else {})
+            request.send_response(200)
+            request.end_headers()
+        elif request.path == "/v1/table/test/create/?mode=create":
+            request.send_response(200)
+            request.send_header("Content-Type", "application/json")
+            request.end_headers()
+            request.wfile.write(b"{}")
+        elif request.path == "/v1/table/test/describe/":
+            request.send_response(200)
+            request.send_header("Content-Type", "application/json")
+            request.end_headers()
+            request.wfile.write(
+                json.dumps(
+                    dict(
+                        version=1,
+                        schema=dict(
+                            fields=[
+                                dict(name="id", type={"type": "int64"}, nullable=False),
+                                dict(
+                                    name="category",
+                                    type={"type": "string"},
+                                    nullable=False,
+                                ),
+                                dict(
+                                    name="text", type={"type": "string"}, nullable=False
+                                ),
+                                dict(
+                                    name="vector",
+                                    type={
+                                        "type": "fixed_size_list",
+                                        "fields": [
+                                            dict(
+                                                name="item",
+                                                type={"type": "float"},
+                                                nullable=True,
+                                            )
+                                        ],
+                                        "length": 2,
+                                    },
+                                    nullable=False,
+                                ),
+                            ]
+                        ),
+                    )
+                ).encode()
+            )
+        else:
+            request.send_response(404)
+            request.end_headers()
+
+    from lancedb.index import BTree, FTS, IvfPq, IvfRq
+
+    with mock_lancedb_connection(handler) as db:
+        table = db.create_table("test", [{"id": 1}])
+
+        # New API: column-first, config= kwarg. Should NOT emit DeprecationWarning.
+        import warnings as _warnings
+
+        with _warnings.catch_warnings():
+            _warnings.simplefilter("error", DeprecationWarning)
+            table.create_index("vector", config=IvfPq(distance_type="l2"))
+            table.create_index("category", config=BTree())
+            table.create_index("text", config=FTS())
+            # IvfRq via new API
+            table.create_index("vector", config=IvfRq(distance_type="l2"))
+
+        # Legacy index_type="IVF_RQ" routes to IvfRq config under the hood.
+        with pytest.warns(DeprecationWarning, match="create_index"):
+            table.create_index(
+                vector_column_name="vector",
+                index_type="IVF_RQ",
+                num_partitions=8,
+            )
+
+        assert len(received_requests) == 5
+        assert [req["column"] for req in received_requests] == [
+            "vector",
+            "category",
+            "text",
+            "vector",
+            "vector",
+        ]
+
+
 def test_table_wait_for_index_timeout():
     def handler(request):
         index_stats = dict(
@@ -1289,6 +1550,10 @@ def _remote_fork_child(port: int, queue) -> None:
     queue.put(db.table_names())
 
 
+def _remote_table_fork_child(table, queue) -> None:
+    queue.put(table.count_rows())
+
+
 @pytest.mark.skipif(
     sys.platform != "linux",
     reason=(
@@ -1351,3 +1616,65 @@ def test_remote_connection_after_fork():
     finally:
         server.shutdown()
         server_thread.join()
+
+
+@pytest.mark.skipif(
+    sys.platform != "linux",
+    reason=(
+        "fork() is unavailable on Windows and unsafe on macOS "
+        "(Apple frameworks/TLS are not fork-safe)"
+    ),
+)
+def test_inherited_remote_table_reopens_after_fork():
+    def handler(request):
+        if request.path == "/v1/table/test/describe/":
+            request.send_response(200)
+            request.send_header("Content-Type", "application/json")
+            request.end_headers()
+            request.wfile.write(b'{"version": 1, "schema": {"fields": []}}')
+        elif request.path == "/v1/table/test/count_rows/":
+            request.send_response(200)
+            request.send_header("Content-Type", "application/json")
+            request.end_headers()
+            request.wfile.write(b"7")
+        else:
+            request.send_response(404)
+            request.end_headers()
+
+    server = http.server.HTTPServer(("localhost", 0), make_mock_http_handler(handler))
+    port = server.server_address[1]
+    server_thread = threading.Thread(target=server.serve_forever)
+    server_thread.start()
+    try:
+        db = lancedb.connect(
+            "db://dev",
+            api_key="fake",
+            host_override=f"http://localhost:{port}",
+            client_config={
+                "retry_config": {"retries": 0},
+                "timeout_config": {"connect_timeout": 2, "read_timeout": 2},
+            },
+        )
+        table = db.open_table("test")
+        assert table.count_rows() == 7
+
+        ctx = mp.get_context("fork")
+        queue = ctx.Queue()
+        proc = ctx.Process(target=_remote_table_fork_child, args=(table, queue))
+        proc.start()
+        proc.join(timeout=15)
+
+        if proc.is_alive():
+            proc.terminate()
+            proc.join(timeout=5)
+            if proc.is_alive():
+                proc.kill()
+                proc.join()
+            pytest.fail("Remote table hung after fork")
+
+        assert proc.exitcode == 0, f"child exited with code {proc.exitcode}"
+        assert not queue.empty(), "child produced no result"
+        assert queue.get() == 7
+    finally:
+        server.shutdown()
+        server_thread.join()

python/python/tests/test_rerankers.py

@@ -603,3 +603,89 @@ def test_cross_encoder_reranker_return_all(tmp_path):
     assert "_relevance_score" in result.column_names
     assert "_score" in result.column_names
     assert "_distance" in result.column_names
+
+
+# ---------------------------------------------------------------------------
+# Regression tests for LinearCombinationReranker scoring bugs (issue #3154)
+# ---------------------------------------------------------------------------
+
+
+def test_linear_combination_best_match_ranks_first():
+    """
+    The document that is BOTH the closest vector match AND the only FTS match
+    must rank first.  Previously _combine_score subtracted from 1, inverting
+    the ranking so the worst document ranked highest.
+    """
+    reranker = LinearCombinationReranker(weight=0.7, return_score="all")
+
+    # rowid 0: perfect vector match, sole FTS match  → should rank 1st
+    # rowid 1: mediocre vector, no FTS match
+    # rowid 2: bad vector, no FTS match
+    vector_results = pa.Table.from_pydict(
+        {
+            "_rowid": [0, 1, 2],
+            "_distance": [0.0, 0.5, 0.9],
+        }
+    )
+    fts_results = pa.Table.from_pydict(
+        {
+            "_rowid": [0],
+            "_score": [1.0],
+        }
+    )
+
+    combined = reranker.merge_results(vector_results, fts_results, fill=1.0)
+    scores = dict(
+        zip(
+            combined["_rowid"].to_pylist(),
+            combined["_relevance_score"].to_pylist(),
+        )
+    )
+
+    # rowid 0 must have the highest relevance score
+    assert scores[0] > scores[1], (
+        f"Best match (rowid 0, score={scores[0]:.4f}) should beat "
+        f"mid match (rowid 1, score={scores[1]:.4f})"
+    )
+    assert scores[1] > scores[2], (
+        f"Mid match (rowid 1, score={scores[1]:.4f}) should beat "
+        f"bad match (rowid 2, score={scores[2]:.4f})"
+    )
+
+
+def test_linear_combination_missing_fts_is_penalised():
+    """
+    A document with no FTS match must score *lower* than a document that
+    has a mediocre FTS match, everything else being equal.  Previously
+    missing-FTS entries used fill=1.0 directly, which gave them a reward
+    (via the 1-(...) inversion) instead of a penalty.
+    """
+    reranker = LinearCombinationReranker(weight=0.5, return_score="all")
+
+    vector_results = pa.Table.from_pydict(
+        {
+            "_rowid": [0, 1],
+            "_distance": [0.2, 0.2],  # identical vector scores
+        }
+    )
+    fts_results = pa.Table.from_pydict(
+        {
+            "_rowid": [0],  # rowid 1 has no FTS match
+            "_score": [0.3],  # small FTS score
+        }
+    )
+
+    combined = reranker.merge_results(vector_results, fts_results, fill=1.0)
+    scores = dict(
+        zip(
+            combined["_rowid"].to_pylist(),
+            combined["_relevance_score"].to_pylist(),
+        )
+    )
+
+    # rowid 0 has a small FTS score; rowid 1 has none.
+    # Even a small FTS contribution should beat having none at all.
+    assert scores[0] > scores[1], (
+        f"Document with FTS score (rowid 0, {scores[0]:.4f}) should beat "
+        f"document with no FTS match (rowid 1, {scores[1]:.4f})"
+    )

python/python/tests/test_table.py

@@ -4,6 +4,7 @@
 
 import os
 import sys
+import warnings
 from datetime import date, datetime, timedelta
 from time import sleep
 from typing import List
@@ -11,7 +12,7 @@ from unittest.mock import patch
 
 import lancedb
 from lancedb.dependencies import _PANDAS_AVAILABLE
-from lancedb.index import HnswFlat, HnswPq, HnswSq, IvfPq
+from lancedb.index import BTree, FTS, HnswFlat, HnswPq, HnswSq, IvfPq
 import numpy as np
 import polars as pl
 import pyarrow as pa
@@ -33,7 +34,7 @@ def test_basic(mem_db: DBConnection):
     table = mem_db.create_table("test", data=data)
 
     assert table.name == "test"
-    assert "LanceTable(name='test', version=1, _conn=LanceDBConnection(" in repr(table)
+    assert "LanceTable(name='test', _conn=LanceDBConnection(" in repr(table)
     expected_schema = pa.schema(
         {
             "vector": pa.list_(pa.float32(), 2),
@@ -928,7 +929,12 @@ def test_create_index_method(mock_create_index, mem_db: DBConnection):
         num_bits=4,
     )
     mock_create_index.assert_called_with(
-        "vector", replace=True, config=expected_config, name=None, train=True
+        "vector",
+        replace=True,
+        config=expected_config,
+        wait_timeout=None,
+        name=None,
+        train=True,
     )
 
     # Test with target_partition_size
@@ -948,7 +954,12 @@ def test_create_index_method(mock_create_index, mem_db: DBConnection):
         target_partition_size=8192,
     )
     mock_create_index.assert_called_with(
-        "vector", replace=True, config=expected_config, name=None, train=True
+        "vector",
+        replace=True,
+        config=expected_config,
+        wait_timeout=None,
+        name=None,
+        train=True,
     )
 
     # target_partition_size has a default value,
@@ -967,7 +978,12 @@ def test_create_index_method(mock_create_index, mem_db: DBConnection):
         num_bits=4,
     )
     mock_create_index.assert_called_with(
-        "vector", replace=True, config=expected_config, name=None, train=True
+        "vector",
+        replace=True,
+        config=expected_config,
+        wait_timeout=None,
+        name=None,
+        train=True,
     )
 
     table.create_index(
@@ -978,7 +994,12 @@ def test_create_index_method(mock_create_index, mem_db: DBConnection):
     )
     expected_config = HnswPq(distance_type="dot")
     mock_create_index.assert_called_with(
-        "my_vector", replace=False, config=expected_config, name=None, train=True
+        "my_vector",
+        replace=False,
+        config=expected_config,
+        wait_timeout=None,
+        name=None,
+        train=True,
     )
 
     table.create_index(
@@ -993,7 +1014,12 @@ def test_create_index_method(mock_create_index, mem_db: DBConnection):
         distance_type="cosine", sample_rate=0.1, m=29, ef_construction=10
     )
     mock_create_index.assert_called_with(
-        "my_vector", replace=True, config=expected_config, name=None, train=True
+        "my_vector",
+        replace=True,
+        config=expected_config,
+        wait_timeout=None,
+        name=None,
+        train=True,
     )
 
     table.create_index(
@@ -1008,7 +1034,12 @@ def test_create_index_method(mock_create_index, mem_db: DBConnection):
         distance_type="cosine", sample_rate=0.1, m=29, ef_construction=10
     )
     mock_create_index.assert_called_with(
-        "my_vector", replace=True, config=expected_config, name=None, train=True
+        "my_vector",
+        replace=True,
+        config=expected_config,
+        wait_timeout=None,
+        name=None,
+        train=True,
     )
 
 
@@ -1032,6 +1063,7 @@ def test_create_index_name_and_train_parameters(
         "vector",
         replace=True,
         config=expected_config,
+        wait_timeout=None,
         name="my_custom_index",
         train=True,
     )
@@ -1039,13 +1071,82 @@ def test_create_index_name_and_train_parameters(
     # Test with train=False
     table.create_index(vector_column_name="vector", train=False)
     mock_create_index.assert_called_with(
-        "vector", replace=True, config=expected_config, name=None, train=False
+        "vector",
+        replace=True,
+        config=expected_config,
+        wait_timeout=None,
+        name=None,
+        train=False,
     )
 
     # Test with both name and train
     table.create_index(vector_column_name="vector", name="my_index_name", train=True)
     mock_create_index.assert_called_with(
-        "vector", replace=True, config=expected_config, name="my_index_name", train=True
+        "vector",
+        replace=True,
+        config=expected_config,
+        wait_timeout=None,
+        name="my_index_name",
+        train=True,
+    )
+
+
+@patch("lancedb.table.AsyncTable.create_index")
+def test_create_index_legacy_emits_deprecation_warning(
+    mock_create_index, mem_db: DBConnection
+):
+    table = mem_db.create_table(
+        "test",
+        data=[{"vector": [3.1, 4.1]}, {"vector": [5.9, 26.5]}],
+    )
+
+    with pytest.warns(DeprecationWarning, match="create_index"):
+        table.create_index(metric="l2", num_partitions=8, vector_column_name="vector")
+
+
+@patch("lancedb.table.AsyncTable.create_index")
+def test_create_index_new_api(mock_create_index, mem_db: DBConnection):
+    table = mem_db.create_table(
+        "test",
+        data=[
+            {"vector": [3.1, 4.1], "category": "a", "text": "hello world"},
+            {"vector": [5.9, 26.5], "category": "b", "text": "goodbye"},
+        ],
+    )
+
+    # Vector index via new API should not warn
+    with warnings.catch_warnings():
+        warnings.simplefilter("error", DeprecationWarning)
+        table.create_index("vector", config=IvfPq(distance_type="l2"))
+    mock_create_index.assert_called_with(
+        "vector",
+        replace=True,
+        config=IvfPq(distance_type="l2"),
+        wait_timeout=None,
+        name=None,
+        train=True,
+    )
+
+    # Scalar index via new API
+    table.create_index("category", config=BTree())
+    mock_create_index.assert_called_with(
+        "category",
+        replace=True,
+        config=BTree(),
+        wait_timeout=None,
+        name=None,
+        train=True,
+    )
+
+    # FTS index via new API
+    table.create_index("text", config=FTS(with_position=True))
+    mock_create_index.assert_called_with(
+        "text",
+        replace=True,
+        config=FTS(with_position=True),
+        wait_timeout=None,
+        name=None,
+        train=True,
     )
 
 
@@ -1861,8 +1962,9 @@ def test_create_scalar_index(mem_db: DBConnection):
         "my_table",
         data=test_data,
     )
-    # Test with default name
-    table.create_scalar_index("x")
+    # Test with default name; confirm DeprecationWarning fires
+    with pytest.warns(DeprecationWarning, match="create_scalar_index"):
+        table.create_scalar_index("x")
     indices = table.list_indices()
     assert len(indices) == 1
     scalar_index = indices[0]
@@ -1934,6 +2036,10 @@ def test_create_index_nested_field_paths(mem_db: DBConnection):
     assert len(vector_results) == 1
     assert vector_results[0]["metadata"]["user_id"] == 0
 
+    default_vector_results = table.search([0.0, 1.0]).limit(1).to_list()
+    assert len(default_vector_results) == 1
+    assert default_vector_results[0]["metadata"]["user_id"] == 0
+
     filtered_results = table.search().where("metadata.user_id = 42").limit(1).to_list()
     assert len(filtered_results) == 1
     assert filtered_results[0]["metadata"]["user_id"] == 42
@@ -2013,6 +2119,74 @@ def test_search_with_schema_inf_multiple_vector(mem_db: DBConnection):
         table.search(q).limit(1).to_arrow()
 
 
+def test_search_infers_single_nested_vector(mem_db: DBConnection):
+    schema = pa.schema(
+        [
+            pa.field("id", pa.int32()),
+            pa.field(
+                "image",
+                pa.struct([pa.field("embedding", pa.list_(pa.float32(), 2))]),
+            ),
+        ]
+    )
+    data = pa.Table.from_pylist(
+        [
+            {"id": 0, "image": {"embedding": [0.0, 1.0]}},
+            {"id": 1, "image": {"embedding": [10.0, 11.0]}},
+        ],
+        schema=schema,
+    )
+    table = mem_db.create_table("nested_vector_default_search", data=data)
+
+    result = table.search([0.0, 1.0]).limit(1).to_list()
+    assert result[0]["id"] == 0
+
+
+def test_search_nested_vector_multiple_candidates(mem_db: DBConnection):
+    schema = pa.schema(
+        [
+            pa.field(
+                "image",
+                pa.struct([pa.field("embedding", pa.list_(pa.float32(), 2))]),
+            ),
+            pa.field(
+                "text",
+                pa.struct([pa.field("embedding", pa.list_(pa.float32(), 2))]),
+            ),
+        ]
+    )
+    data = pa.Table.from_pylist(
+        [
+            {
+                "image": {"embedding": [0.0, 1.0]},
+                "text": {"embedding": [2.0, 3.0]},
+            }
+        ],
+        schema=schema,
+    )
+    table = mem_db.create_table("nested_vector_multiple_candidates", data=data)
+
+    with pytest.raises(ValueError, match="image.embedding.*text.embedding"):
+        table.search([0.0, 1.0]).limit(1).to_arrow()
+
+
+def test_search_nested_vector_no_candidates(mem_db: DBConnection):
+    schema = pa.schema(
+        [
+            pa.field("id", pa.int32()),
+            pa.field("metadata", pa.struct([pa.field("label", pa.string())])),
+        ]
+    )
+    data = pa.Table.from_pylist(
+        [{"id": 0, "metadata": {"label": "cat"}}],
+        schema=schema,
+    )
+    table = mem_db.create_table("nested_vector_no_candidates", data=data)
+
+    with pytest.raises(ValueError, match="no vector column"):
+        table.search([0.0, 1.0]).limit(1).to_arrow()
+
+
 def test_compact_cleanup(tmp_db: DBConnection):
     pytest.importorskip("lance")
     table = tmp_db.create_table(
@@ -2298,6 +2472,30 @@ def test_alter_columns(mem_db: DBConnection):
     assert table.to_arrow().column_names == ["new_id"]
 
 
+def test_update_field_metadata(mem_db: DBConnection):
+    data = pa.table({"id": [0, 1], "category": ["a", "b"]})
+    table = mem_db.create_table("my_table", data=data)
+
+    res = table.update_field_metadata(
+        {"path": "category", "metadata": {"unit": "label", "pii": "false"}}
+    )
+    assert res.version == 2
+    # Arrow field metadata is bytes-keyed
+    assert table.schema.field("category").metadata == {
+        b"unit": b"label",
+        b"pii": b"false",
+    }
+
+    # merge: add a key, delete one via None, keep the rest
+    table.update_field_metadata(
+        {"path": "category", "metadata": {"source": "import", "pii": None}}
+    )
+    assert table.schema.field("category").metadata == {
+        b"unit": b"label",
+        b"source": b"import",
+    }
+
+
 @pytest.mark.asyncio
 async def test_alter_columns_async(mem_db_async: AsyncConnection):
     data = pa.table({"id": [0, 1]})

python/python/tests/test_torch.py

@@ -1,10 +1,15 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright The LanceDB Authors
 
+import contextlib
 import functools
+import http.server
+import json
 import multiprocessing as mp
 import pickle
+import re
 import sys
+import threading
 
 import lancedb
 import pyarrow as pa
@@ -15,6 +20,107 @@ from lancedb.util import tbl_to_tensor
 torch = pytest.importorskip("torch")
 
 
+REMOTE_ROWS = list(range(100))
+
+
+def _make_mock_http_handler(handler):
+    class MockLanceDBHandler(http.server.BaseHTTPRequestHandler):
+        def do_GET(self):
+            handler(self)
+
+        def do_POST(self):
+            handler(self)
+
+    return MockLanceDBHandler
+
+
+def _remote_schema_payload():
+    return {
+        "version": 1,
+        "schema": {
+            "fields": [
+                {"name": "a", "type": {"type": "int64"}, "nullable": False},
+            ]
+        },
+    }
+
+
+def _offsets_from_filter(filter_sql: str | None) -> list[int]:
+    if filter_sql is None:
+        return REMOTE_ROWS
+    match = re.search(r"_rowoffset in \((.*?)\)", filter_sql)
+    if match is None:
+        return REMOTE_ROWS
+    raw_offsets = match.group(1).strip()
+    if raw_offsets == "":
+        return []
+    return [int(offset.strip()) for offset in raw_offsets.split(",")]
+
+
+def _remote_dataset_handler(request):
+    request.close_connection = True
+    if request.path == "/v1/table/test/describe/":
+        request.send_response(200)
+        request.send_header("Content-Type", "application/json")
+        request.end_headers()
+        request.wfile.write(json.dumps(_remote_schema_payload()).encode())
+    elif request.path == "/v1/table/test/count_rows/":
+        request.send_response(200)
+        request.send_header("Content-Type", "application/json")
+        request.end_headers()
+        request.wfile.write(str(len(REMOTE_ROWS)).encode())
+    elif request.path == "/v1/table/test/query/":
+        content_len = int(request.headers.get("Content-Length"))
+        body = json.loads(request.rfile.read(content_len))
+        offsets = _offsets_from_filter(body.get("filter"))
+        requested_columns = body.get("columns") or ["a"]
+        if isinstance(requested_columns, dict):
+            requested_columns = list(requested_columns)
+
+        data = {}
+        for column in requested_columns:
+            if column == "a":
+                data[column] = [REMOTE_ROWS[offset] for offset in offsets]
+            elif column == "_rowoffset":
+                data[column] = offsets
+            elif column == "_rowid":
+                data[column] = offsets
+
+        table = pa.table(data)
+        request.send_response(200)
+        request.send_header("Content-Type", "application/vnd.apache.arrow.file")
+        request.end_headers()
+        with pa.ipc.new_file(request.wfile, schema=table.schema) as writer:
+            writer.write_table(table)
+    else:
+        request.send_response(404)
+        request.end_headers()
+
+
+@contextlib.contextmanager
+def _remote_dataset_table():
+    with http.server.ThreadingHTTPServer(
+        ("localhost", 0), _make_mock_http_handler(_remote_dataset_handler)
+    ) as server:
+        port = server.server_address[1]
+        handle = threading.Thread(target=server.serve_forever)
+        handle.start()
+        try:
+            db = lancedb.connect(
+                "db://dev",
+                api_key="fake",
+                host_override=f"http://localhost:{port}",
+                client_config={
+                    "retry_config": {"retries": 0},
+                    "timeout_config": {"connect_timeout": 2, "read_timeout": 2},
+                },
+            )
+            yield db.open_table("test")
+        finally:
+            server.shutdown()
+            handle.join()
+
+
 def _open_native_table(uri: str, table_name: str):
     """Top-level connection factory used by the explicit-factory pickle test.
 
@@ -107,6 +213,39 @@ def test_permutation_dataloader_multiprocessing(tmp_db):
     assert seen == 1000
 
 
+def test_remote_table_dataloader_multiprocessing():
+    with _remote_dataset_table() as table:
+        dataloader = torch.utils.data.DataLoader(
+            table,
+            collate_fn=tbl_to_tensor,
+            batch_size=10,
+            num_workers=2,
+            multiprocessing_context="spawn",
+        )
+        seen = 0
+        for batch in dataloader:
+            assert batch.size(0) == 1
+            assert batch.size(1) == 10
+            seen += batch.size(1)
+        assert seen == len(REMOTE_ROWS)
+
+
+def test_remote_permutation_dataloader_multiprocessing():
+    with _remote_dataset_table() as table:
+        permutation = Permutation.identity(table)
+        dataloader = torch.utils.data.DataLoader(
+            permutation,
+            batch_size=10,
+            num_workers=2,
+            multiprocessing_context="spawn",
+        )
+        seen = 0
+        for batch in dataloader:
+            assert batch["a"].size(0) == 10
+            seen += batch["a"].size(0)
+        assert seen == len(REMOTE_ROWS)
+
+
 def test_permutation_pickle_with_connection_factory(tmp_path):
     """When the user provides a connection_factory, pickling should round-trip
     through that factory rather than introspecting the connection URI. Useful
@@ -171,6 +310,35 @@ def _multiworker_dataloader_target(db_uri: str, result_queue):
     result_queue.put(count)
 
 
+def _remote_multiworker_dataloader_target(port: int, result_queue):
+    import lancedb
+    from lancedb.permutation import Permutation
+
+    db = lancedb.connect(
+        "db://dev",
+        api_key="fake",
+        host_override=f"http://localhost:{port}",
+        client_config={
+            "retry_config": {"retries": 0},
+            "timeout_config": {"connect_timeout": 2, "read_timeout": 2},
+        },
+    )
+    table = db.open_table("test")
+    permutation = Permutation.identity(table)
+
+    dataloader = torch.utils.data.DataLoader(
+        permutation,
+        batch_size=10,
+        num_workers=2,
+        multiprocessing_context="fork",
+    )
+    count = 0
+    for batch in dataloader:
+        assert batch["a"].size(0) == 10
+        count += 1
+    result_queue.put(count)
+
+
 @pytest.mark.skipif(
     sys.platform != "linux",
     reason=(
@@ -208,3 +376,46 @@ def test_permutation_dataloader_fork_workers(tmp_path):
     assert proc.exitcode == 0, f"child exited with code {proc.exitcode}"
     assert not queue.empty(), "child produced no batches"
     assert queue.get() == 100
+
+
+@pytest.mark.skipif(
+    sys.platform != "linux",
+    reason=(
+        "fork() is unavailable on Windows and unsafe on macOS "
+        "(Apple frameworks/TLS are not fork-safe)"
+    ),
+)
+def test_remote_permutation_dataloader_fork_workers():
+    with http.server.ThreadingHTTPServer(
+        ("localhost", 0), _make_mock_http_handler(_remote_dataset_handler)
+    ) as server:
+        port = server.server_address[1]
+        handle = threading.Thread(target=server.serve_forever)
+        handle.start()
+        try:
+            ctx = mp.get_context("spawn")
+            queue = ctx.Queue()
+            proc = ctx.Process(
+                target=_remote_multiworker_dataloader_target,
+                args=(port, queue),
+            )
+            proc.start()
+            proc.join(timeout=30)
+
+            if proc.is_alive():
+                proc.terminate()
+                proc.join(timeout=5)
+                if proc.is_alive():
+                    proc.kill()
+                    proc.join()
+                pytest.fail(
+                    "Remote permutation hung when iterated in a fork-based "
+                    "DataLoader worker"
+                )
+
+            assert proc.exitcode == 0, f"child exited with code {proc.exitcode}"
+            assert not queue.empty(), "child produced no batches"
+            assert queue.get() == 10
+        finally:
+            server.shutdown()
+            handle.join()

python/src/lib.rs

@@ -16,7 +16,7 @@ use query::{FTSQuery, HybridQuery, Query, VectorQuery};
 use session::Session;
 use table::{
     AddColumnsResult, AddResult, AlterColumnsResult, DeleteResult, DropColumnsResult, LsmWriteSpec,
-    MergeResult, Table, UpdateResult,
+    MergeResult, Table, UpdateFieldMetadataResult, UpdateResult,
 };
 
 pub mod arrow;
@@ -50,6 +50,7 @@ pub fn _lancedb(_py: Python, m: &Bound<'_, PyModule>) -> PyResult<()> {
     m.add_class::<RecordBatchStream>()?;
     m.add_class::<AddColumnsResult>()?;
     m.add_class::<AlterColumnsResult>()?;
+    m.add_class::<UpdateFieldMetadataResult>()?;
     m.add_class::<AddResult>()?;
     m.add_class::<MergeResult>()?;
     m.add_class::<LsmWriteSpec>()?;

python/src/table.rs

@@ -16,12 +16,12 @@ use arrow::{
     pyarrow::{FromPyArrow, PyArrowType, ToPyArrow},
 };
 use lancedb::table::{
-    AddDataMode, ColumnAlteration, Duration, NewColumnTransform, OptimizeAction, OptimizeOptions,
-    Table as LanceDbTable,
+    AddDataMode, ColumnAlteration, Duration, FieldMetadataUpdate, NewColumnTransform,
+    OptimizeAction, OptimizeOptions, Table as LanceDbTable,
 };
 use pyo3::{
     Bound, FromPyObject, Py, PyAny, PyRef, PyResult, Python,
-    exceptions::{PyKeyError, PyRuntimeError, PyValueError},
+    exceptions::{PyRuntimeError, PyValueError},
     pyclass, pymethods,
     types::{IntoPyDict, PyAnyMethods, PyDict, PyDictMethods},
 };
@@ -143,18 +143,20 @@ pub struct MergeResult {
     pub num_inserted_rows: u64,
     pub num_deleted_rows: u64,
     pub num_attempts: u32,
+    pub num_rows: u64,
 }
 
 #[pymethods]
 impl MergeResult {
     pub fn __repr__(&self) -> String {
         format!(
-            "MergeResult(version={}, num_updated_rows={}, num_inserted_rows={}, num_deleted_rows={}, num_attempts={})",
+            "MergeResult(version={}, num_updated_rows={}, num_inserted_rows={}, num_deleted_rows={}, num_attempts={}, num_rows={})",
             self.version,
             self.num_updated_rows,
             self.num_inserted_rows,
             self.num_deleted_rows,
-            self.num_attempts
+            self.num_attempts,
+            self.num_rows
         )
     }
 }
@@ -167,6 +169,7 @@ impl From<lancedb::table::MergeResult> for MergeResult {
             num_inserted_rows: result.num_inserted_rows,
             num_deleted_rows: result.num_deleted_rows,
             num_attempts: result.num_attempts,
+            num_rows: result.num_rows,
         }
     }
 }
@@ -194,6 +197,12 @@ impl LsmWriteSpec {
     }
 
     /// Identity sharding — shard by the raw value of `column`.
+    ///
+    /// `column` must be a deterministic function of the unenforced primary
+    /// key: every row with a given primary key must always produce the same
+    /// `column` value, or upserts of that key can land in different shards
+    /// and a stale version can win. Typically `column` is the primary key
+    /// itself or a stable attribute of it.
     #[staticmethod]
     pub fn identity(column: String) -> Self {
         Self {
@@ -348,6 +357,27 @@ impl From<lancedb::table::AlterColumnsResult> for AlterColumnsResult {
     }
 }
 
+#[pyclass(get_all, from_py_object)]
+#[derive(Clone, Debug)]
+pub struct UpdateFieldMetadataResult {
+    pub version: u64,
+}
+
+#[pymethods]
+impl UpdateFieldMetadataResult {
+    pub fn __repr__(&self) -> String {
+        format!("UpdateFieldMetadataResult(version={})", self.version)
+    }
+}
+
+impl From<lancedb::table::UpdateFieldMetadataResult> for UpdateFieldMetadataResult {
+    fn from(result: lancedb::table::UpdateFieldMetadataResult) -> Self {
+        Self {
+            version: result.version,
+        }
+    }
+}
+
 #[pyclass(get_all, from_py_object)]
 #[derive(Clone, Debug)]
 pub struct DropColumnsResult {
@@ -933,6 +963,12 @@ impl Table {
         if let Some(use_index) = parameters.use_index {
             builder.use_index(use_index);
         }
+        if let Some(use_lsm_write) = parameters.use_lsm_write {
+            builder.use_lsm_write(use_lsm_write);
+        }
+        if let Some(validate_single_shard) = parameters.validate_single_shard {
+            builder.validate_single_shard(validate_single_shard);
+        }
 
         future_into_py(self_.py(), async move {
             let res = builder.execute(Box::new(batches)).await.infer_error()?;
@@ -971,6 +1007,13 @@ impl Table {
         })
     }
 
+    pub fn close_lsm_writers(self_: PyRef<'_, Self>) -> PyResult<Bound<'_, PyAny>> {
+        let inner = self_.inner_ref()?.clone();
+        future_into_py(self_.py(), async move {
+            inner.close_lsm_writers().await.infer_error()
+        })
+    }
+
     pub fn uses_v2_manifest_paths(self_: PyRef<'_, Self>) -> PyResult<Bound<'_, PyAny>> {
         let inner = self_.inner_ref()?.clone();
         future_into_py(self_.py(), async move {
@@ -1080,31 +1123,57 @@ impl Table {
         field_name: String,
         metadata: &Bound<'_, PyDict>,
     ) -> PyResult<Bound<'a, PyAny>> {
-        let mut new_metadata = HashMap::<String, String>::new();
-        for (column_name, value) in metadata.into_iter() {
-            let key: String = column_name.extract()?;
-            let value: String = value.extract()?;
-            new_metadata.insert(key, value);
+        // Deprecated: forwards to the update_field_metadata path (replace mode).
+        let mut update = FieldMetadataUpdate::new(field_name).replace();
+        for (key, value) in metadata.into_iter() {
+            update = update.set(key.extract::<String>()?, value.extract::<String>()?);
         }
 
         let inner = self_.inner_ref()?.clone();
         future_into_py(self_.py(), async move {
-            let native_tbl = inner
-                .as_native()
-                .ok_or_else(|| PyValueError::new_err("This cannot be run on a remote table"))?;
-            let schema = native_tbl.manifest().await.infer_error()?.schema;
-            let field = schema
-                .field(&field_name)
-                .ok_or_else(|| PyKeyError::new_err(format!("Field {} not found", field_name)))?;
-
-            native_tbl
-                .replace_field_metadata(vec![(field.id as u32, new_metadata)])
-                .await
-                .infer_error()?;
-
+            inner.update_field_metadata(&[update]).await.infer_error()?;
             Ok(())
         })
     }
+
+    pub fn update_field_metadata<'a>(
+        self_: PyRef<'a, Self>,
+        updates: Vec<Bound<PyDict>>,
+    ) -> PyResult<Bound<'a, PyAny>> {
+        let updates = updates
+            .iter()
+            .map(|update| {
+                let path: String = update
+                    .get_item("path")?
+                    .ok_or_else(|| PyValueError::new_err("Missing path"))?
+                    .extract()?;
+                let mut field_update = FieldMetadataUpdate::new(path);
+                if let Some(metadata) = update.get_item("metadata")? {
+                    let metadata_dict = metadata.cast::<PyDict>()?;
+                    for (key, value) in metadata_dict.iter() {
+                        let key: String = key.extract()?;
+                        if value.is_none() {
+                            field_update = field_update.remove(key);
+                        } else {
+                            field_update = field_update.set(key, value.extract::<String>()?);
+                        }
+                    }
+                }
+                if let Some(replace) = update.get_item("replace")?
+                    && replace.extract::<bool>()?
+                {
+                    field_update = field_update.replace();
+                }
+                Ok(field_update)
+            })
+            .collect::<PyResult<Vec<_>>>()?;
+
+        let inner = self_.inner_ref()?.clone();
+        future_into_py(self_.py(), async move {
+            let result = inner.update_field_metadata(&updates).await.infer_error()?;
+            Ok(UpdateFieldMetadataResult::from(result))
+        })
+    }
 }
 
 #[derive(FromPyObject)]
@@ -1124,6 +1193,8 @@ pub struct MergeInsertParams {
     when_not_matched_by_source_condition: Option<String>,
     timeout: Option<std::time::Duration>,
     use_index: Option<bool>,
+    use_lsm_write: Option<bool>,
+    validate_single_shard: Option<bool>,
 }
 
 #[pyclass]

rust-toolchain.toml

@@ -1,2 +1,2 @@
 [toolchain]
-channel = "1.94.0"
+channel = "1.95.0"

rust/lancedb/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "lancedb"
-version = "0.29.1-beta.0"
+version = "0.30.1-beta.0"
 edition.workspace = true
 description = "LanceDB: A serverless, low-latency vector database for AI applications"
 license.workspace = true
@@ -75,7 +75,7 @@ reqwest = { version = "0.12.0", default-features = false, features = [
     "stream",
 ], optional = true }
 http = { version = "1", optional = true } # Matching what is in reqwest
-uuid = { version = "1.7.0", features = ["v4"] }
+uuid = { version = "1.7.0", features = ["v4", "v5"] }
 polars-arrow = { version = ">=0.37,<0.40.0", optional = true }
 polars = { version = ">=0.37,<0.40.0", optional = true }
 hf-hub = { version = "0.4.1", optional = true, default-features = false, features = [
@@ -104,6 +104,7 @@ datafusion.workspace = true
 http-body = "1"                                        # Matching reqwest
 rstest = "0.23.0"
 test-log = "0.2"
+serial_test = "3"
 
 
 [features]

rust/lancedb/src/connection.rs

@@ -812,8 +812,7 @@ impl ConnectBuilder {
         self
     }
 
-    /// The interval at which to check for updates from other processes. This
-    /// only affects LanceDB OSS.
+    /// The interval at which to check for updates from other processes.
     ///
     /// If left unset, consistency is not checked. For maximum read
     /// performance, this is the default. For strong consistency, set this to
@@ -825,8 +824,11 @@ impl ConnectBuilder {
     /// This only affects read operations. Write operations are always
     /// consistent.
     ///
-    /// LanceDB Cloud uses eventual consistency under the hood, and is not
-    /// currently configurable.
+    /// # Cost
+    ///
+    /// Stronger consistency is not free. The smaller the interval, the more
+    /// often each read pays the cost of checking for updates against object
+    /// storage, raising per-read latency and cost.
     pub fn read_consistency_interval(
         mut self,
         read_consistency_interval: std::time::Duration,
@@ -886,6 +888,7 @@ impl ConnectBuilder {
             options.host_override,
             self.request.client_config,
             storage_options.into(),
+            self.request.read_consistency_interval,
         )?);
         Ok(Connection {
             internal,

rust/lancedb/src/dataloader/permutation/shuffle.rs

@@ -464,11 +464,9 @@ mod tests {
         let mut iter = ids.into_iter().map(|o| o.unwrap());
         while let Some(first) = iter.next() {
             let rows_left_in_clump = if first == 4470 { 19 } else { 29 };
-            let mut expected_next = first + 1;
-            for _ in 0..rows_left_in_clump {
+            for expected_next in (first + 1)..=(first + rows_left_in_clump) {
                 let next = iter.next().unwrap();
                 assert_eq!(next, expected_next);
-                expected_next += 1;
             }
         }
     }

rust/lancedb/src/index/vector.rs

@@ -23,17 +23,12 @@ impl VectorIndex {
             .fields
             .iter()
             .map(|field_id| {
-                manifest
-                    .schema
-                    .field_by_id(*field_id)
-                    .unwrap_or_else(|| {
-                        panic!(
-                            "field {field_id} of index {} must exist in schema",
-                            index.name
-                        )
-                    })
-                    .name
-                    .clone()
+                manifest.schema.field_path(*field_id).unwrap_or_else(|_| {
+                    panic!(
+                        "field {field_id} of index {} must exist in schema",
+                        index.name
+                    )
+                })
             })
             .collect();
         Self {

rust/lancedb/src/remote/client.rs

@@ -245,6 +245,9 @@ pub struct RestfulLanceDbClient<S: HttpSend = Sender> {
     pub(crate) sender: S,
     pub(crate) id_delimiter: String,
     pub(crate) header_provider: Option<Arc<dyn HeaderProvider>>,
+    /// Connection-level read consistency interval. Drives the
+    /// `x-lancedb-min-timestamp` freshness header sent on read requests.
+    pub(crate) read_consistency_interval: Option<Duration>,
 }
 
 impl<S: HttpSend> std::fmt::Debug for RestfulLanceDbClient<S> {
@@ -338,6 +341,7 @@ impl RestfulLanceDbClient<Sender> {
         host_override: Option<String>,
         default_headers: HeaderMap,
         client_config: ClientConfig,
+        read_consistency_interval: Option<Duration>,
     ) -> Result<Self> {
         // Get the timeouts
         let timeout =
@@ -435,6 +439,7 @@ impl RestfulLanceDbClient<Sender> {
                 .clone()
                 .unwrap_or("$".to_string()),
             header_provider: client_config.header_provider,
+            read_consistency_interval,
         })
     }
 }
@@ -840,6 +845,16 @@ pub mod test_utils {
     pub fn client_with_handler<T>(
         handler: impl Fn(reqwest::Request) -> http::response::Response<T> + Send + Sync + 'static,
     ) -> RestfulLanceDbClient<MockSender>
+    where
+        T: Into<reqwest::Body>,
+    {
+        client_with_handler_and_interval(handler, None)
+    }
+
+    pub fn client_with_handler_and_interval<T>(
+        handler: impl Fn(reqwest::Request) -> http::response::Response<T> + Send + Sync + 'static,
+        read_consistency_interval: Option<Duration>,
+    ) -> RestfulLanceDbClient<MockSender>
     where
         T: Into<reqwest::Body>,
     {
@@ -857,6 +872,7 @@ pub mod test_utils {
             },
             id_delimiter: "$".to_string(),
             header_provider: None,
+            read_consistency_interval,
         }
     }
 
@@ -881,6 +897,7 @@ pub mod test_utils {
             },
             id_delimiter: config.id_delimiter.unwrap_or_else(|| "$".to_string()),
             header_provider: config.header_provider,
+            read_consistency_interval: None,
         }
     }
 }
@@ -888,8 +905,18 @@ pub mod test_utils {
 #[cfg(test)]
 mod tests {
     use super::*;
+    use serial_test::serial;
     use std::time::Duration;
 
+    // Serializes the env-var-mutating tests below: cargo test runs tests in
+    // parallel, but several of these tests read and write the same process-
+    // global env vars (`LANCEDB_USER_ID*`), so they would race without this.
+    static ENV_MUTEX: std::sync::Mutex<()> = std::sync::Mutex::new(());
+
+    fn lock_env() -> std::sync::MutexGuard<'static, ()> {
+        ENV_MUTEX.lock().unwrap_or_else(|e| e.into_inner())
+    }
+
     #[test]
     fn test_timeout_config_default() {
         let config = TimeoutConfig::default();
@@ -1046,6 +1073,7 @@ mod tests {
             sender: Sender,
             id_delimiter: "+".to_string(),
             header_provider: Some(Arc::new(provider) as Arc<dyn HeaderProvider>),
+            read_consistency_interval: None,
         };
 
         // Apply dynamic headers
@@ -1081,6 +1109,7 @@ mod tests {
             sender: Sender,
             id_delimiter: "+".to_string(),
             header_provider: Some(Arc::new(provider) as Arc<dyn HeaderProvider>),
+            read_consistency_interval: None,
         };
 
         // Apply dynamic headers
@@ -1118,6 +1147,7 @@ mod tests {
             sender: Sender,
             id_delimiter: "+".to_string(),
             header_provider: Some(Arc::new(provider) as Arc<dyn HeaderProvider>),
+            read_consistency_interval: None,
         };
 
         // Header provider errors should fail the request
@@ -1143,7 +1173,9 @@ mod tests {
     }
 
     #[test]
+    #[serial(user_id_env)]
     fn test_resolve_user_id_none() {
+        let _guard = lock_env();
         let config = ClientConfig::default();
         // Clear env vars that might be set from other tests
         // SAFETY: This is only called in tests
@@ -1155,7 +1187,9 @@ mod tests {
     }
 
     #[test]
+    #[serial(user_id_env)]
     fn test_resolve_user_id_from_env() {
+        let _guard = lock_env();
         // SAFETY: This is only called in tests
         unsafe {
             std::env::set_var("LANCEDB_USER_ID", "env-user-id");
@@ -1169,7 +1203,9 @@ mod tests {
     }
 
     #[test]
+    #[serial(user_id_env)]
     fn test_resolve_user_id_from_env_key() {
+        let _guard = lock_env();
         // SAFETY: This is only called in tests
         unsafe {
             std::env::remove_var("LANCEDB_USER_ID");
@@ -1189,7 +1225,9 @@ mod tests {
     }
 
     #[test]
+    #[serial(user_id_env)]
     fn test_resolve_user_id_direct_takes_precedence() {
+        let _guard = lock_env();
         // SAFETY: This is only called in tests
         unsafe {
             std::env::set_var("LANCEDB_USER_ID", "env-user-id");
@@ -1206,7 +1244,9 @@ mod tests {
     }
 
     #[test]
+    #[serial(user_id_env)]
     fn test_resolve_user_id_empty_env_ignored() {
+        let _guard = lock_env();
         // SAFETY: This is only called in tests
         unsafe {
             std::env::set_var("LANCEDB_USER_ID", "");

rust/lancedb/src/remote/db.rs

@@ -206,6 +206,7 @@ impl RemoteDatabase {
         host_override: Option<String>,
         client_config: ClientConfig,
         options: RemoteOptions,
+        read_consistency_interval: Option<std::time::Duration>,
     ) -> Result<Self> {
         let parsed = super::client::parse_db_url(uri)?;
         let header_map = RestfulLanceDbClient::<Sender>::default_headers(
@@ -233,6 +234,7 @@ impl RemoteDatabase {
             host_override,
             header_map,
             client_config.clone(),
+            read_consistency_interval,
         )?;
 
         let table_cache = Cache::builder()

rust/lancedb/src/table.rs

@@ -89,10 +89,12 @@ use futures::future::join_all;
 pub use lance::dataset::refs::{TagContents, Tags as LanceTags};
 pub use lance::dataset::scanner::DatasetRecordBatchStream;
 use lance::dataset::statistics::DatasetStatisticsExt;
-use lance_index::frag_reuse::FRAG_REUSE_INDEX_NAME;
 pub use lance_index::optimize::OptimizeOptions;
 pub use optimize::{CompactionOptions, OptimizeAction, OptimizeStats};
-pub use schema_evolution::{AddColumnsResult, AlterColumnsResult, DropColumnsResult};
+pub use schema_evolution::{
+    AddColumnsResult, AlterColumnsResult, DropColumnsResult, FieldMetadataUpdate,
+    UpdateFieldMetadataResult,
+};
 use serde_with::skip_serializing_none;
 pub use update::{UpdateBuilder, UpdateResult};
 
@@ -253,6 +255,36 @@ pub enum Filter {
     Datafusion(Expr),
 }
 
+/// A predicate for filtering rows in delete operations.
+///
+/// Accepts either a SQL string or a DataFusion [`Expr`]. Use the [`From`]
+/// implementations to convert from `&str` or `&Expr` automatically.
+/// See [`Table::delete`] for usage examples.
+pub enum Predicate<'a> {
+    /// A SQL predicate string
+    String(&'a str),
+    /// A DataFusion logical expression
+    Expr(&'a Expr),
+}
+
+impl<'a> From<&'a str> for Predicate<'a> {
+    fn from(s: &'a str) -> Self {
+        Predicate::String(s)
+    }
+}
+
+impl<'a> From<&'a String> for Predicate<'a> {
+    fn from(s: &'a String) -> Self {
+        Predicate::String(s.as_str())
+    }
+}
+
+impl<'a> From<&'a Expr> for Predicate<'a> {
+    fn from(e: &'a Expr) -> Self {
+        Predicate::Expr(e)
+    }
+}
+
 #[async_trait]
 pub trait Tags: Send + Sync {
     /// List the tags of the table.
@@ -282,17 +314,15 @@ pub use self::merge::MergeResult;
 /// date) and [`LsmWriteSpec::with_writer_config_defaults`] (default
 /// `ShardWriter` configuration recorded in the MemWAL index).
 ///
-/// All variants require the table to have an unenforced primary key.
-///
 /// Install a spec with [`Table::set_lsm_write_spec`] and remove it with
 /// [`Table::unset_lsm_write_spec`]. The actual `merge_insert` dispatch
 /// onto the MemWAL writer is a follow-up.
 #[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
 pub enum LsmWriteSpec {
-    /// Hash-bucket sharding by the unenforced primary key column.
+    /// Hash-bucket sharding by a scalar column.
     ///
-    /// `column` must equal the table's currently-set single-column
-    /// unenforced primary key. `num_buckets` must be in `[1, 1024]`.
+    /// `column` must be a non-nested column with a supported scalar type.
+    /// `num_buckets` must be in `[1, 1024]`.
     /// Iceberg-compatible Murmur3-x86-32 (seed 0) is used so each row's
     /// `bucket(column, num_buckets)` value is stable across processes.
     Bucket {
@@ -339,6 +369,14 @@ impl LsmWriteSpec {
 
     /// Construct an identity-sharding spec (shard by the raw value of
     /// `column`) with no maintained indexes.
+    ///
+    /// `column` must be a deterministic function of the unenforced primary
+    /// key: every row with a given primary key must always produce the same
+    /// `column` value. MemWAL dedups upserts by primary key but tracks
+    /// generations per shard, so if the same key is written with two
+    /// different `column` values its versions land in different shards and a
+    /// stale value can win. Typically `column` is the primary key itself, or
+    /// a stable attribute of it (e.g. a tenant id).
     pub fn identity(column: impl Into<String>) -> Self {
         Self::Identity {
             column: column.into(),
@@ -491,8 +529,8 @@ pub trait BaseTable: std::fmt::Display + std::fmt::Debug + Send + Sync {
 
     /// Add new records to the table.
     async fn add(&self, add: AddDataBuilder) -> Result<AddResult>;
-    /// Delete rows from the table.
-    async fn delete(&self, predicate: &str) -> Result<DeleteResult>;
+    /// Delete rows from the table matching the given [`Predicate`].
+    async fn delete(&self, predicate: Predicate<'_>) -> Result<DeleteResult>;
     /// Update rows in the table.
     async fn update(&self, update: UpdateBuilder) -> Result<UpdateResult>;
     /// Create an index on the provided column(s).
@@ -553,6 +591,13 @@ pub trait BaseTable: std::fmt::Display + std::fmt::Debug + Send + Sync {
             message: "unset_lsm_write_spec is not supported on this table type".into(),
         })
     }
+    /// Drain and close any cached MemWAL shard writers for this table.
+    ///
+    /// The default implementation is a no-op; table types that maintain
+    /// MemWAL shard writers override it.
+    async fn close_lsm_writers(&self) -> Result<()> {
+        Ok(())
+    }
     /// Gets the table tag manager.
     async fn tags(&self) -> Result<Box<dyn Tags + '_>>;
     /// Optimize the dataset.
@@ -618,6 +663,19 @@ pub trait BaseTable: std::fmt::Display + std::fmt::Debug + Send + Sync {
             message: "create_insert_exec not implemented".to_string(),
         })
     }
+    /// Update per-field metadata. Merges into existing metadata by default;
+    /// [`FieldMetadataUpdate::remove`] deletes a key and
+    /// [`FieldMetadataUpdate::replace`] swaps the field's whole map.
+    ///
+    /// The default returns `NotSupported`; Lance-backed and remote tables override it.
+    async fn update_field_metadata(
+        &self,
+        _updates: &[FieldMetadataUpdate],
+    ) -> Result<UpdateFieldMetadataResult> {
+        Err(Error::NotSupported {
+            message: "update_field_metadata is not supported on this table type".into(),
+        })
+    }
 }
 
 /// A Table is a collection of strong typed Rows.
@@ -656,6 +714,30 @@ mod test_utils {
             }
         }
 
+        pub fn new_with_handler_and_interval<T>(
+            name: impl Into<String>,
+            handler: impl Fn(reqwest::Request) -> http::Response<T> + Clone + Send + Sync + 'static,
+            read_consistency_interval: Option<std::time::Duration>,
+        ) -> Self
+        where
+            T: Into<reqwest::Body>,
+        {
+            let inner = Arc::new(
+                crate::remote::table::RemoteTable::new_mock_with_consistency_interval(
+                    name.into(),
+                    handler.clone(),
+                    read_consistency_interval,
+                ),
+            );
+            let database = Arc::new(crate::remote::db::RemoteDatabase::new_mock(handler));
+            Self {
+                inner,
+                database: Some(database),
+                // Registry is unused.
+                embedding_registry: Arc::new(MemoryRegistry::new()),
+            }
+        }
+
         pub fn new_with_handler_version<T>(
             name: impl Into<String>,
             version: semver::Version,
@@ -860,7 +942,8 @@ impl Table {
     /// Delete the rows from table that match the predicate.
     ///
     /// # Arguments
-    /// - `predicate` - The SQL predicate string to filter the rows to be deleted.
+    /// - `predicate` - A SQL string (`&str`) or DataFusion expression (`&Expr`)
+    ///   that selects the rows to delete.
     ///
     /// # Example
     ///
@@ -869,6 +952,7 @@ impl Table {
     /// # use arrow_array::{FixedSizeListArray, types::Float32Type, RecordBatch,
     /// #   RecordBatchIterator, Int32Array};
     /// # use arrow_schema::{Schema, Field, DataType};
+    /// use datafusion_expr::{col, lit};
     /// # tokio::runtime::Runtime::new().unwrap().block_on(async {
     /// let tmpdir = tempfile::tempdir().unwrap();
     /// let db = lancedb::connect(tmpdir.path().to_str().unwrap())
@@ -898,11 +982,17 @@ impl Table {
     ///     .execute()
     ///     .await
     ///     .unwrap();
+    ///
+    /// // Using a SQL string:
     /// tbl.delete("id > 5").await.unwrap();
+    ///
+    /// // Using a DataFusion expression:
+    /// let expr = col("id").lt(lit(4));
+    /// tbl.delete(&expr).await.unwrap();
     /// # });
     /// ```
-    pub async fn delete(&self, predicate: &str) -> Result<DeleteResult> {
-        self.inner.delete(predicate).await
+    pub async fn delete(&self, predicate: impl Into<Predicate<'_>>) -> Result<DeleteResult> {
+        self.inner.delete(predicate.into()).await
     }
 
     /// Create an index on the provided column(s).
@@ -1266,6 +1356,14 @@ impl Table {
         self.inner.alter_columns(alterations).await
     }
 
+    /// Update per-field metadata (merges by default).
+    pub async fn update_field_metadata(
+        &self,
+        updates: &[FieldMetadataUpdate],
+    ) -> Result<UpdateFieldMetadataResult> {
+        self.inner.update_field_metadata(updates).await
+    }
+
     /// Remove columns from the table.
     pub async fn drop_columns(&self, columns: &[&str]) -> Result<DropColumnsResult> {
         self.inner.drop_columns(columns).await
@@ -1298,21 +1396,15 @@ impl Table {
     ///
     /// [`LsmWriteSpec`] chooses one of three sharding strategies:
     ///
-    /// - [`LsmWriteSpec::bucket`] — hash-bucket writes by the single-column
-    ///   unenforced primary key.
+    /// - [`LsmWriteSpec::bucket`] — hash-bucket writes by a scalar column.
     /// - [`LsmWriteSpec::identity`] — shard by the raw value of a scalar column.
     /// - [`LsmWriteSpec::unsharded`] — route every write to a single shard.
     ///
-    /// All variants require the table to have an unenforced primary key
-    /// ([`Table::set_unenforced_primary_key`]); bucket sharding additionally
-    /// requires it to be the single column being bucketed.
-    ///
     /// # Example
     ///
     /// ```
     /// # use lancedb::table::{LsmWriteSpec, Table};
     /// # async fn example(table: &Table) -> Result<(), Box<dyn std::error::Error>> {
-    /// table.set_unenforced_primary_key(["id"]).await?;
     /// table
     ///     .set_lsm_write_spec(
     ///         LsmWriteSpec::bucket("id", 16).with_maintained_indexes(["id_idx"]),
@@ -1333,6 +1425,16 @@ impl Table {
         self.inner.unset_lsm_write_spec().await
     }
 
+    /// Drain and close any cached MemWAL shard writers held for this table.
+    ///
+    /// When an [`LsmWriteSpec`] is installed, `merge_insert` opens MemWAL shard
+    /// writers and caches them for reuse across calls. This closes them,
+    /// flushing pending data; writers reopen lazily on the next `merge_insert`.
+    /// It is a no-op when no writers are cached.
+    pub async fn close_lsm_writers(&self) -> Result<()> {
+        self.inner.close_lsm_writers().await
+    }
+
     /// Retrieve the version of the table
     ///
     /// LanceDb supports versioning.  Every operation that modifies the table increases
@@ -2502,6 +2604,7 @@ impl NativeTable {
     ///   field id and the second element is a hashmap of metadata key-value
     ///   pairs.
     ///
+    #[deprecated(since = "0.33.1", note = "Use `update_field_metadata` instead")]
     pub async fn replace_field_metadata(
         &self,
         new_values: impl IntoIterator<Item = (u32, HashMap<String, String>)>,
@@ -2688,16 +2791,13 @@ impl BaseTable for NativeTable {
                 message: "Multi-column (composite) indices are not yet supported".to_string(),
             });
         }
-
-        let dataset = self.dataset.get().await?;
+        self.dataset.ensure_mutable()?;
+        let mut dataset = (*self.dataset.get().await?).clone();
         let (column, field) = Self::resolve_index_field(dataset.schema(), &opts.columns[0])?;
-        drop(dataset);
 
         let lance_idx_params = self.make_index_params(&field, opts.index.clone()).await?;
         let index_type = self.get_index_type_for_field(&field, &opts.index);
         let columns = [column.as_str()];
-        self.dataset.ensure_mutable()?;
-        let mut dataset = (*self.dataset.get().await?).clone();
         let mut builder = dataset
             .create_index_builder(&columns, index_type, lance_idx_params.as_ref())
             .train(opts.train)
@@ -2779,9 +2879,12 @@ impl BaseTable for NativeTable {
         merge::lsm::unset_lsm_write_spec(self).await
     }
 
+    async fn close_lsm_writers(&self) -> Result<()> {
+        merge::lsm::close_lsm_writers(self).await
+    }
+
     /// Delete rows from the table
-    async fn delete(&self, predicate: &str) -> Result<DeleteResult> {
-        // Delegate to the submodule implementation
+    async fn delete(&self, predicate: Predicate<'_>) -> Result<DeleteResult> {
         delete::execute_delete(self, predicate).await
     }
 
@@ -2808,72 +2911,62 @@ impl BaseTable for NativeTable {
         schema_evolution::execute_alter_columns(self, alterations).await
     }
 
+    async fn update_field_metadata(
+        &self,
+        updates: &[FieldMetadataUpdate],
+    ) -> Result<UpdateFieldMetadataResult> {
+        schema_evolution::execute_update_field_metadata(self, updates).await
+    }
+
     async fn drop_columns(&self, columns: &[&str]) -> Result<DropColumnsResult> {
         schema_evolution::execute_drop_columns(self, columns).await
     }
 
     async fn list_indices(&self) -> Result<Vec<IndexConfig>> {
         let dataset = self.dataset.get().await?;
-        let indices = dataset.load_indices().await?;
-        let results = futures::stream::iter(indices.as_slice()).then(|idx| async {
-
-            // skip Lance internal indexes
-            if idx.name == FRAG_REUSE_INDEX_NAME {
-                return None;
-            }
-
-            let stats = match dataset.index_statistics(idx.name.as_str()).await {
-                Ok(stats) => stats,
-                Err(e) => {
-                    log::warn!("Failed to get statistics for index {} ({}): {}", idx.name, idx.uuid, e);
-                    return None;
-                }
-            };
-
-            let stats: serde_json::Value = match serde_json::from_str(&stats) {
-                Ok(stats) => stats,
-                Err(e) => {
-                    log::warn!("Failed to deserialize index statistics for index {} ({}): {}", idx.name, idx.uuid, e);
-                    return None;
-                }
-            };
-
-            let Some(index_type) = stats.get("index_type").and_then(|v| v.as_str()) else {
-                log::warn!("Index statistics was missing 'index_type' field for index {} ({})", idx.name, idx.uuid);
-                return None;
-            };
-
-            let index_type: crate::index::IndexType = match index_type.parse() {
-                Ok(index_type) => index_type,
-                Err(e) => {
-                    log::warn!("Failed to parse index type for index {} ({}): {}", idx.name, idx.uuid, e);
-                    return None;
-                }
-            };
-
-            let mut columns = Vec::with_capacity(idx.fields.len());
-            for field_id in &idx.fields {
-                let column = match dataset.schema().field_path(*field_id) {
-                    Ok(column) => column,
+        let indices = dataset
+            .describe_indices(None)
+            .await?
+            .into_iter()
+            .filter_map(|idx_desc| {
+                let index_type: crate::index::IndexType = match idx_desc.index_type().parse() {
+                    Ok(index_type) => index_type,
                     Err(e) => {
                         log::warn!(
-                            "The index {} ({}) referenced a field with id {} which does not exist in the schema: {}",
-                            idx.name,
-                            idx.uuid,
-                            field_id,
+                            "Failed to parse index type for index {}: {}",
+                            idx_desc.name(),
                             e
                         );
                         return None;
                     }
                 };
-                columns.push(column);
-            }
 
-            let name = idx.name.clone();
-            Some(IndexConfig { index_type, columns, name })
-        }).collect::<Vec<_>>().await;
+                let field_ids = idx_desc.field_ids();
+                let mut columns = Vec::with_capacity(field_ids.len());
+                for field_id in field_ids {
+                    let field_path = match dataset.schema().field_path(*field_id as i32) {
+                        Ok(field_path) => field_path,
+                        Err(e) => {
+                            log::warn!(
+                                "Failed to resolve field path for index {} field id {}: {}",
+                                idx_desc.name(),
+                                field_id,
+                                e
+                            );
+                            return None;
+                        }
+                    };
+                    columns.push(field_path);
+                }
 
-        Ok(results.into_iter().flatten().collect())
+                Some(IndexConfig {
+                    name: idx_desc.name().to_string(),
+                    index_type,
+                    columns,
+                })
+            })
+            .collect();
+        Ok(indices)
     }
 
     async fn uri(&self) -> Result<String> {
@@ -2983,11 +3076,12 @@ impl BaseTable for NativeTable {
         let p99 = *sorted_sizes.get(num_fragments * 99 / 100).unwrap_or(&0);
         let min = sorted_sizes.first().copied().unwrap_or(0);
         let max = sorted_sizes.last().copied().unwrap_or(0);
-        let mean = if num_fragments == 0 {
-            0
-        } else {
-            sorted_sizes.iter().copied().sum::<usize>() / num_fragments
-        };
+        let mean = sorted_sizes
+            .iter()
+            .copied()
+            .sum::<usize>()
+            .checked_div(num_fragments)
+            .unwrap_or(0);
 
         let frag_stats = FragmentStatistics {
             num_fragments,
@@ -3854,6 +3948,25 @@ mod tests {
             1
         );
 
+        let default_vector_results = table
+            .query()
+            .nearest_to(&[0.0; 8])
+            .unwrap()
+            .limit(1)
+            .execute()
+            .await
+            .unwrap()
+            .try_collect::<Vec<_>>()
+            .await
+            .unwrap();
+        assert_eq!(
+            default_vector_results
+                .iter()
+                .map(|batch| batch.num_rows())
+                .sum::<usize>(),
+            1
+        );
+
         let fts_results = table
             .query()
             .full_text_search(FullTextSearchQuery::new("document".to_string()))
@@ -3967,26 +4080,27 @@ mod tests {
         let index_configs = table.list_indices().await.unwrap();
         assert_eq!(index_configs.len(), 5);
 
+        // list_indices returns indices in alphabetical order by name
         let mut configs_iter = index_configs.into_iter();
         let index = configs_iter.next().unwrap();
         assert_eq!(index.index_type, crate::index::IndexType::Bitmap);
         assert_eq!(index.columns, vec!["category".to_string()]);
 
-        let index = configs_iter.next().unwrap();
-        assert_eq!(index.index_type, crate::index::IndexType::Bitmap);
-        assert_eq!(index.columns, vec!["is_active".to_string()]);
-
         let index = configs_iter.next().unwrap();
         assert_eq!(index.index_type, crate::index::IndexType::Bitmap);
         assert_eq!(index.columns, vec!["data".to_string()]);
 
         let index = configs_iter.next().unwrap();
         assert_eq!(index.index_type, crate::index::IndexType::Bitmap);
-        assert_eq!(index.columns, vec!["large_data".to_string()]);
+        assert_eq!(index.columns, vec!["is_active".to_string()]);
 
         let index = configs_iter.next().unwrap();
         assert_eq!(index.index_type, crate::index::IndexType::Bitmap);
         assert_eq!(index.columns, vec!["large_category".to_string()]);
+
+        let index = configs_iter.next().unwrap();
+        assert_eq!(index.index_type, crate::index::IndexType::Bitmap);
+        assert_eq!(index.columns, vec!["large_data".to_string()]);
     }
 
     #[tokio::test]
@@ -4366,10 +4480,10 @@ mod tests {
             Some(&"test_val2_update".to_string())
         );
 
-        let mut new_field_metadata = HashMap::<String, String>::new();
-        new_field_metadata.insert("test_field_key1".into(), "test_field_val1".into());
         native_tbl
-            .replace_field_metadata(vec![(field.id as u32, new_field_metadata)])
+            .update_field_metadata(&[
+                FieldMetadataUpdate::new("i").set("test_field_key1", "test_field_val1")
+            ])
             .await
             .unwrap();
 
@@ -4558,21 +4672,6 @@ mod tests {
             .unwrap();
         let table = conn.create_table("t", reader).execute().await.unwrap();
 
-        // Reject when no PK is set.
-        let err = table
-            .set_lsm_write_spec(LsmWriteSpec::bucket("id", 4))
-            .await
-            .expect_err("should reject without PK");
-        assert!(matches!(err, Error::Lance { .. }), "got {:?}", err);
-
-        // Set PK, then a mismatched column on the spec must be rejected.
-        table.set_unenforced_primary_key(["id"]).await.unwrap();
-        let err = table
-            .set_lsm_write_spec(LsmWriteSpec::bucket("name", 4))
-            .await
-            .expect_err("should reject column != PK");
-        assert!(matches!(err, Error::Lance { .. }), "got {:?}", err);
-
         // Reject num_buckets out of range.
         for bad in [0u32, 1025] {
             let err = table
@@ -4638,9 +4737,6 @@ mod tests {
             .unwrap();
         let table = conn.create_table("t", reader).execute().await.unwrap();
 
-        // Lance's MemWAL still requires *some* unenforced primary key on
-        // the dataset; Unsharded just skips the per-row hashing step.
-        table.set_unenforced_primary_key(["id"]).await.unwrap();
         table
             .set_lsm_write_spec(LsmWriteSpec::unsharded())
             .await
@@ -4687,7 +4783,6 @@ mod tests {
             .unwrap();
         let table = conn.create_table("t", reader).execute().await.unwrap();
 
-        table.set_unenforced_primary_key(["id"]).await.unwrap();
         table
             .set_lsm_write_spec(
                 LsmWriteSpec::identity("region")
@@ -4743,7 +4838,6 @@ mod tests {
         table.unset_lsm_write_spec().await.unwrap_err();
 
         // Install a spec, then unset it.
-        table.set_unenforced_primary_key(["id"]).await.unwrap();
         table
             .set_lsm_write_spec(LsmWriteSpec::bucket("id", 4))
             .await

rust/lancedb/src/table/add_data.rs

@@ -982,4 +982,105 @@ mod tests {
         table2.add(struct_batch).execute().await.unwrap();
         assert_eq!(table2.count_rows(None).await.unwrap(), 2);
     }
+
+    /// Regression test: appending `arrow.json` (PyArrow `pa.json_()`) data into a table
+    /// whose schema was created with `pa.json_()` (internally stored as `lance.json`, backed
+    /// by `LargeBinary`) must succeed without a schema-mismatch error.
+    ///
+    /// Previously `build_field_exprs` would attempt a `Utf8 → LargeBinary` DataFusion cast,
+    /// which produced a field whose Arrow extension metadata still read `arrow.json` instead
+    /// of `lance.json`.  Lance-core then rejected the append with
+    /// `"json vs large_binary" schema mismatch`.
+    ///
+    /// PyArrow's `pa.json_()` may be backed by either `Utf8` or `LargeUtf8` depending on the
+    /// constructor used, so the test is parameterized over the input backing type.
+    #[rstest::rstest]
+    #[case::utf8(DataType::Utf8)]
+    #[case::large_utf8(DataType::LargeUtf8)]
+    #[tokio::test]
+    async fn test_add_arrow_json_into_lance_json_table(#[case] input_type: DataType) {
+        use arrow_array::{Array, cast::AsArray};
+        use lance_arrow::ARROW_EXT_NAME_KEY;
+        use lance_arrow::json::{ARROW_JSON_EXT_NAME, JSON_EXT_NAME};
+
+        // Build a table whose "data" column is lance.json (LargeBinary +
+        // ARROW:extension:name = "lance.json").
+        let lance_json_field = lance_arrow::json::json_field("data", true);
+        let table_schema = Arc::new(Schema::new(vec![lance_json_field]));
+
+        let db = connect("memory://").execute().await.unwrap();
+        let table = db
+            .create_empty_table("json_test", table_schema)
+            .execute()
+            .await
+            .unwrap();
+
+        // Sanity-check the stored schema.
+        let stored_field = table.schema().await.unwrap();
+        let data_field = stored_field.field_with_name("data").unwrap();
+        assert_eq!(data_field.data_type(), &DataType::LargeBinary);
+        assert_eq!(
+            data_field
+                .metadata()
+                .get(ARROW_EXT_NAME_KEY)
+                .map(|s| s.as_str()),
+            Some(JSON_EXT_NAME),
+        );
+
+        // Build an arrow.json input field (Utf8/LargeUtf8 + arrow.json extension).
+        // This is what PyArrow produces for pa.json_() arrays.
+        let arrow_json_metadata = std::collections::HashMap::from([(
+            ARROW_EXT_NAME_KEY.to_string(),
+            ARROW_JSON_EXT_NAME.to_string(),
+        )]);
+        let arrow_json_field =
+            Field::new("data", input_type.clone(), true).with_metadata(arrow_json_metadata);
+        let arrow_json_schema = Arc::new(Schema::new(vec![arrow_json_field]));
+
+        let rows: Vec<Option<&str>> = vec![None, Some(r#"{"a": 1}"#), Some(r#"{"b": 2}"#)];
+        let string_array: Arc<dyn arrow_array::Array> = match input_type {
+            DataType::Utf8 => Arc::new(arrow_array::StringArray::from(rows.clone())),
+            DataType::LargeUtf8 => Arc::new(arrow_array::LargeStringArray::from(rows.clone())),
+            other => panic!("unsupported arrow.json backing type for this test: {other:?}"),
+        };
+        let batch = RecordBatch::try_new(arrow_json_schema, vec![string_array]).unwrap();
+
+        // This must not fail with a schema-mismatch error.
+        table.add(batch).execute().await.unwrap();
+
+        assert_eq!(table.count_rows(None).await.unwrap(), rows.len());
+
+        // A lance.json column is read back as Utf8 carrying arrow.json extension metadata.
+        let results: Vec<RecordBatch> = table
+            .query()
+            .select(Select::columns(&["data"]))
+            .execute()
+            .await
+            .unwrap()
+            .try_collect()
+            .await
+            .unwrap();
+
+        assert_eq!(results.len(), 1);
+        let batch = &results[0];
+        assert_eq!(batch.num_rows(), rows.len());
+
+        let json_col = batch.column(0);
+        assert_eq!(json_col.data_type(), &DataType::Utf8);
+        let json_strs = json_col.as_string::<i32>();
+
+        for (i, expected) in rows.iter().enumerate() {
+            match expected {
+                None => assert!(json_strs.is_null(i), "row {i} expected null"),
+                Some(raw) => {
+                    assert!(!json_strs.is_null(i), "row {i} expected non-null");
+                    let actual: serde_json::Value = serde_json::from_str(json_strs.value(i))
+                        .expect("read-back JSON should be valid");
+                    let expected: serde_json::Value =
+                        serde_json::from_str(raw).expect("expected JSON should be valid");
+                    assert_eq!(actual, expected, "row {i} JSON mismatch");
+                }
+            }
+        }
+    }
 }

rust/lancedb/src/table/datafusion/cast.rs

@@ -13,6 +13,7 @@ use datafusion_physical_expr::expressions::{CastExpr, Literal};
 use datafusion_physical_plan::expressions::Column;
 use datafusion_physical_plan::projection::ProjectionExec;
 use datafusion_physical_plan::{ExecutionPlan, PhysicalExpr};
+use lance_arrow::json::{is_arrow_json_field, is_json_field};
 
 use crate::{Error, Result};
 
@@ -64,6 +65,18 @@ fn build_field_exprs(
         let input_field = &input_fields[input_idx];
         let input_expr = get_input_expr(input_idx);
 
+        // Special case: input is arrow.json (PyArrow pa.json_() extension type backed by
+        // Utf8/LargeUtf8) and the table field is lance.json (backed by LargeBinary).
+        // Lance-core's write path already handles the arrow.json → lance.json conversion
+        // (including JSONB encoding), so we pass the expression through unchanged and let
+        // lance-core deal with it. Attempting to cast Utf8 → LargeBinary here would
+        // produce a field whose metadata still identifies it as arrow.json, which then
+        // causes a schema-mismatch error inside lance-core.
+        if is_arrow_json_field(input_field) && is_json_field(table_field) {
+            result.push((input_expr, Arc::clone(input_field) as FieldRef));
+            continue;
+        }
+
         let expr = match (input_field.data_type(), table_field.data_type()) {
             // Both are structs: recurse into sub-fields to handle subschemas and casts.
             (DataType::Struct(in_children), DataType::Struct(tbl_children))
@@ -618,4 +631,75 @@ mod tests {
             .unwrap();
         assert_eq!(a.values(), &[1, 3]);
     }
+
+    /// `arrow.json` input (PyArrow `pa.json_()`, Utf8/LargeUtf8 + extension metadata) against a
+    /// `lance.json` table field (LargeBinary + extension metadata) must be passed through
+    /// without a cast so that lance-core can perform its own arrow.json → JSONB conversion.
+    ///
+    /// Before the fix, `cast_to_table_schema` attempted a `Utf8 → LargeBinary` DataFusion
+    /// cast that preserved the wrong extension metadata, causing lance-core to reject the
+    /// batch with a "json vs large_binary" schema-mismatch error.
+    #[rstest::rstest]
+    #[case::utf8(DataType::Utf8)]
+    #[case::large_utf8(DataType::LargeUtf8)]
+    #[tokio::test]
+    async fn test_arrow_json_passthrough_to_lance_json(#[case] input_type: DataType) {
+        use lance_arrow::ARROW_EXT_NAME_KEY;
+        use lance_arrow::json::{ARROW_JSON_EXT_NAME, json_field};
+
+        // Build a table schema with a lance.json field (LargeBinary + lance.json metadata).
+        let lance_field = json_field("data", true);
+        let table_schema = Schema::new(vec![lance_field]);
+
+        // Build an input batch with an arrow.json field (Utf8/LargeUtf8 + arrow.json metadata).
+        let arrow_meta = std::collections::HashMap::from([(
+            ARROW_EXT_NAME_KEY.to_string(),
+            ARROW_JSON_EXT_NAME.to_string(),
+        )]);
+        let arrow_field = Field::new("data", input_type.clone(), true).with_metadata(arrow_meta);
+        let input_schema = Arc::new(Schema::new(vec![arrow_field]));
+
+        let values = vec![Some(r#"{"x": 1}"#), None, Some(r#"{"y": 2}"#)];
+        let input_array: Arc<dyn arrow_array::Array> = match input_type {
+            DataType::Utf8 => Arc::new(StringArray::from(values)),
+            DataType::LargeUtf8 => Arc::new(arrow_array::LargeStringArray::from(values)),
+            other => panic!("unsupported arrow.json backing type for this test: {other:?}"),
+        };
+        let input_batch = RecordBatch::try_new(input_schema, vec![input_array]).unwrap();
+
+        let plan = plan_from_batch(input_batch).await;
+        let projected = cast_to_table_schema(plan, &table_schema).unwrap();
+
+        // The projected schema's "data" field must carry arrow.json metadata
+        // (the input field), not be silently dropped or miscast.
+        let out_field = projected.schema().field_with_name("data").unwrap().clone();
+        assert_eq!(out_field.data_type(), &input_type);
+        assert_eq!(
+            out_field
+                .metadata()
+                .get(ARROW_EXT_NAME_KEY)
+                .map(|s| s.as_str()),
+            Some(ARROW_JSON_EXT_NAME),
+            "output field must still carry arrow.json metadata so lance-core can handle it"
+        );
+
+        // The data must flow through correctly (3 rows, no panic).
+        let result = collect(projected).await;
+        assert_eq!(result.num_rows(), 3);
+        let (v0, v2) = match input_type {
+            DataType::Utf8 => {
+                let col: &StringArray = result.column(0).as_any().downcast_ref().unwrap();
+                (col.value(0).to_string(), col.value(2).to_string())
+            }
+            DataType::LargeUtf8 => {
+                let col: &arrow_array::LargeStringArray =
+                    result.column(0).as_any().downcast_ref().unwrap();
+                (col.value(0).to_string(), col.value(2).to_string())
+            }
+            _ => unreachable!(),
+        };
+        assert_eq!(v0, r#"{"x": 1}"#);
+        assert!(result.column(0).is_null(1));
+        assert_eq!(v2, r#"{"y": 2}"#);
+    }
 }

rust/lancedb/src/table/dataset.rs

@@ -8,6 +8,7 @@ use std::{
 
 use lance::{Dataset, dataset::refs};
 
+use crate::table::merge::lsm::ShardWriterCache;
 use crate::{Error, error::Result, utils::background_cache::BackgroundCache};
 
 /// A wrapper around a [Dataset] that provides consistency checks.
@@ -18,6 +19,10 @@ use crate::{Error, error::Result, utils::background_cache::BackgroundCache};
 pub struct DatasetConsistencyWrapper {
     state: Arc<Mutex<DatasetState>>,
     consistency: ConsistencyMode,
+    /// The single MemWAL `ShardWriter` for this dataset, co-located so it is
+    /// cached for the session and shares the dataset's lifecycle. A dataset
+    /// writes to one shard at a time. Shared by `Arc` across clones.
+    shard_writer: Arc<ShardWriterCache>,
 }
 
 /// The current dataset and whether it is pinned to a specific version.
@@ -67,9 +72,15 @@ impl DatasetConsistencyWrapper {
                 pinned_version: None,
             })),
             consistency,
+            shard_writer: Arc::new(ShardWriterCache::default()),
         }
     }
 
+    /// The MemWAL `ShardWriter` cache co-located with this dataset.
+    pub(crate) fn shard_writer(&self) -> &Arc<ShardWriterCache> {
+        &self.shard_writer
+    }
+
     /// Get the current dataset.
     ///
     /// Behavior depends on the consistency mode:

rust/lancedb/src/table/delete.rs

@@ -1,9 +1,12 @@
+use std::sync::Arc;
+
 use futures::FutureExt;
+use lance::dataset::DeleteBuilder;
 // SPDX-License-Identifier: Apache-2.0
 // SPDX-FileCopyrightText: Copyright The LanceDB Authors
 use serde::{Deserialize, Serialize};
 
-use super::NativeTable;
+use super::{NativeTable, Predicate};
 use crate::Result;
 
 #[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, Default)]
@@ -21,17 +24,39 @@ pub struct DeleteResult {
 /// Internal implementation of the delete logic
 ///
 /// This logic was moved from NativeTable::delete to keep table.rs clean.
-pub(crate) async fn execute_delete(table: &NativeTable, predicate: &str) -> Result<DeleteResult> {
+pub(crate) async fn execute_delete(
+    table: &NativeTable,
+    predicate: Predicate<'_>,
+) -> Result<DeleteResult> {
     table.dataset.ensure_mutable()?;
-    let mut dataset = (*table.dataset.get().await?).clone();
-    let delete_result = dataset.delete(predicate).boxed().await?;
-    let num_deleted_rows = delete_result.num_deleted_rows;
-    let version = dataset.version().version;
-    table.dataset.update(dataset);
-    Ok(DeleteResult {
-        num_deleted_rows,
-        version,
-    })
+    match predicate {
+        Predicate::String(s) => {
+            let mut dataset = (*table.dataset.get().await?).clone();
+            let delete_result = dataset.delete(s).boxed().await?;
+            let num_deleted_rows = delete_result.num_deleted_rows;
+            let version = dataset.version().version;
+            table.dataset.update(dataset);
+            Ok(DeleteResult {
+                num_deleted_rows,
+                version,
+            })
+        }
+        Predicate::Expr(expr) => {
+            let dataset = table.dataset.get().await?;
+            let delete_result = DeleteBuilder::from_expr(Arc::clone(&dataset), expr.clone())
+                .execute()
+                .await?;
+            let num_deleted_rows = delete_result.num_deleted_rows;
+            let version = delete_result.new_dataset.version().version;
+            table.dataset.update(
+                Arc::try_unwrap(delete_result.new_dataset).unwrap_or_else(|arc| (*arc).clone()),
+            );
+            Ok(DeleteResult {
+                num_deleted_rows,
+                version,
+            })
+        }
+    }
 }
 
 #[cfg(test)]
@@ -176,4 +201,100 @@ mod tests {
             "Table version must increment after delete operation"
         );
     }
+
+    #[tokio::test]
+    async fn test_delete_expr() {
+        use datafusion_expr::{col, lit};
+
+        let conn = connect("memory://").execute().await.unwrap();
+
+        // 1. Create a table with values 0 to 9
+        let schema = Arc::new(Schema::new(vec![Field::new("i", DataType::Int32, false)]));
+        let batch = RecordBatch::try_new(
+            schema.clone(),
+            vec![Arc::new(Int32Array::from_iter_values(0..10))],
+        )
+        .unwrap();
+
+        let table = conn
+            .create_table("test_delete_expr", batch)
+            .execute()
+            .await
+            .unwrap();
+
+        // 2. Verify initial state
+        assert_eq!(table.count_rows(None).await.unwrap(), 10);
+        let initial_version = table.version().await.unwrap();
+
+        // 3. Execute Delete with Expr (removes values > 5)
+        let expr = col("i").gt(lit(5));
+        table.delete(&expr).await.unwrap();
+
+        // 4. Verify results
+        assert_eq!(table.count_rows(None).await.unwrap(), 6); // 0, 1, 2, 3, 4, 5 remain
+        let current_version = table.version().await.unwrap();
+        assert!(
+            current_version > initial_version,
+            "Table version must increment after delete_expr operation"
+        );
+
+        // 5. Verify specific data consistency
+        let batches = table
+            .query()
+            .execute()
+            .await
+            .unwrap()
+            .try_collect::<Vec<_>>()
+            .await
+            .unwrap();
+        let batch = &batches[0];
+        let array = batch
+            .column(0)
+            .as_any()
+            .downcast_ref::<Int32Array>()
+            .unwrap();
+
+        // Ensure no value > 5 exists
+        for val in array.iter() {
+            assert!(val.unwrap() <= 5);
+        }
+    }
+
+    #[tokio::test]
+    async fn test_delete_expr_increments_version() {
+        use datafusion_expr::lit;
+
+        let conn = connect("memory://").execute().await.unwrap();
+
+        // Create a table with 5 rows
+        let batch = record_batch!(("id", Int32, [1, 2, 3, 4, 5])).unwrap();
+
+        let table = conn
+            .create_table("test_delete_expr_noop", batch)
+            .execute()
+            .await
+            .unwrap();
+
+        // Capture the initial state (Rows = 5, Version = 1)
+        let initial_rows = table.count_rows(None).await.unwrap();
+        let initial_version = table.version().await.unwrap();
+
+        assert_eq!(initial_rows, 5);
+        let expr = lit(false);
+        table.delete(&expr).await.unwrap();
+
+        // Rows should still be 5
+        let current_rows = table.count_rows(None).await.unwrap();
+        assert_eq!(
+            current_rows, initial_rows,
+            "Data should not change when predicate is false"
+        );
+
+        // version check
+        let current_version = table.version().await.unwrap();
+        assert!(
+            current_version > initial_version,
+            "Table version must increment after delete_expr operation"
+        );
+    }
 }

rust/lancedb/src/table/merge.rs

@@ -41,6 +41,16 @@ pub struct MergeResult {
     /// A value of 1 means the operation succeeded on the first try.
     #[serde(default)]
     pub num_attempts: u32,
+    /// Total number of rows written.
+    ///
+    /// On the standard `merge_insert` path this equals
+    /// `num_inserted_rows + num_updated_rows`. On the MemWAL LSM write path the
+    /// insert/update breakdown is not known until compaction; in that mode
+    /// `num_inserted_rows`, `num_updated_rows`, `num_deleted_rows`, `version`
+    /// and `num_attempts` are all `0` and this field holds the total number of
+    /// rows written through the shard writer.
+    #[serde(default)]
+    pub num_rows: u64,
 }
 
 /// A builder used to create and run a merge insert operation
@@ -57,6 +67,8 @@ pub struct MergeInsertBuilder {
     pub(crate) when_not_matched_by_source_delete_filt: Option<String>,
     pub(crate) timeout: Option<Duration>,
     pub(crate) use_index: bool,
+    pub(crate) use_lsm_write: Option<bool>,
+    pub(crate) validate_single_shard: bool,
 }
 
 impl MergeInsertBuilder {
@@ -71,6 +83,8 @@ impl MergeInsertBuilder {
             when_not_matched_by_source_delete_filt: None,
             timeout: None,
             use_index: true,
+            use_lsm_write: None,
+            validate_single_shard: true,
         }
     }
 
@@ -150,6 +164,34 @@ impl MergeInsertBuilder {
         self
     }
 
+    /// Controls whether `merge_insert` uses the MemWAL LSM write path.
+    ///
+    /// By default (unset), a `merge_insert` on a table with an
+    /// [`LsmWriteSpec`](super::LsmWriteSpec) installed is routed through
+    /// Lance's MemWAL shard writer, and a table without one uses the standard
+    /// path. Calling this with `false` forces the standard path even when a
+    /// spec is set. Calling it with `true` requires a spec — `merge_insert`
+    /// errors if none is installed.
+    pub fn use_lsm_write(&mut self, use_lsm_write: bool) -> &mut Self {
+        self.use_lsm_write = Some(use_lsm_write);
+        self
+    }
+
+    /// Controls how an LSM `merge_insert` checks that its input targets a
+    /// single shard.
+    ///
+    /// When a table has an LSM write spec, every row in a `merge_insert` call
+    /// must route to the same shard. When `true` (the default), every row is
+    /// inspected to verify this. When `false`, only the first row is inspected
+    /// and the shard it routes to is used for the whole input — a faster path
+    /// for callers that have already pre-sharded their input.
+    ///
+    /// Has no effect on tables without an LSM write spec.
+    pub fn validate_single_shard(&mut self, validate_single_shard: bool) -> &mut Self {
+        self.validate_single_shard = validate_single_shard;
+        self
+    }
+
     /// Executes the merge insert operation
     ///
     /// Returns version and statistics about the merge operation including the number of rows
@@ -167,6 +209,23 @@ pub(crate) async fn execute_merge_insert(
     params: MergeInsertBuilder,
     new_data: Box<dyn RecordBatchReader + Send>,
 ) -> Result<MergeResult> {
+    match lsm::lsm_dispatch_decision(table, &params).await? {
+        lsm::LsmDispatch::Lsm(plan) => {
+            let future =
+                lsm::execute_lsm_merge_insert(table, plan, params.validate_single_shard, new_data);
+            return match params.timeout {
+                Some(timeout) => match tokio::time::timeout(timeout, future).await {
+                    Ok(result) => result,
+                    Err(_) => Err(Error::Runtime {
+                        message: "merge insert timed out".to_string(),
+                    }),
+                },
+                None => future.await,
+            };
+        }
+        lsm::LsmDispatch::Standard => {}
+    }
+
     let dataset = table.dataset.get().await?;
     let mut builder = LanceMergeInsertBuilder::try_new(dataset.clone(), params.on)?;
     match (
@@ -219,6 +278,7 @@ pub(crate) async fn execute_merge_insert(
         num_inserted_rows: stats.num_inserted_rows,
         num_deleted_rows: stats.num_deleted_rows,
         num_attempts: stats.num_attempts,
+        num_rows: stats.num_inserted_rows + stats.num_updated_rows,
     })
 }
 
@@ -327,3 +387,366 @@ mod tests {
         assert_eq!(table.count_rows(None).await.unwrap(), 25);
     }
 }
+
+#[cfg(test)]
+mod lsm_tests {
+    use std::sync::Arc;
+
+    use arrow_array::{
+        Int64Array, RecordBatch, RecordBatchIterator, RecordBatchReader, StringArray,
+    };
+    use arrow_schema::{DataType, Field, Schema};
+    use tempfile::{TempDir, tempdir};
+
+    use crate::connect;
+    use crate::error::Error;
+    use crate::table::{LsmWriteSpec, Table};
+
+    /// A reader of `[id: Int64, value: Int64]` rows; `value` is `0..n`.
+    fn id_value_reader(ids: Vec<i64>) -> Box<dyn RecordBatchReader + Send> {
+        let schema = Arc::new(Schema::new(vec![
+            Field::new("id", DataType::Int64, false),
+            Field::new("value", DataType::Int64, false),
+        ]));
+        let n = ids.len() as i64;
+        let batch = RecordBatch::try_new(
+            schema.clone(),
+            vec![
+                Arc::new(Int64Array::from(ids)),
+                Arc::new(Int64Array::from_iter_values(0..n)),
+            ],
+        )
+        .unwrap();
+        Box::new(RecordBatchIterator::new(vec![Ok(batch)], schema))
+    }
+
+    /// A reader of `[id: Int64, region: Utf8]` rows.
+    fn id_region_reader(rows: Vec<(i64, &str)>) -> Box<dyn RecordBatchReader + Send> {
+        let schema = Arc::new(Schema::new(vec![
+            Field::new("id", DataType::Int64, false),
+            Field::new("region", DataType::Utf8, false),
+        ]));
+        let ids: Vec<i64> = rows.iter().map(|(id, _)| *id).collect();
+        let regions: Vec<&str> = rows.iter().map(|(_, region)| *region).collect();
+        let batch = RecordBatch::try_new(
+            schema.clone(),
+            vec![
+                Arc::new(Int64Array::from(ids)),
+                Arc::new(StringArray::from(regions)),
+            ],
+        )
+        .unwrap();
+        Box::new(RecordBatchIterator::new(vec![Ok(batch)], schema))
+    }
+
+    /// A multi-batch reader of `[id: Int64, region: Utf8]` rows.
+    fn id_region_multi_reader(batches: Vec<Vec<(i64, &str)>>) -> Box<dyn RecordBatchReader + Send> {
+        let schema = Arc::new(Schema::new(vec![
+            Field::new("id", DataType::Int64, false),
+            Field::new("region", DataType::Utf8, false),
+        ]));
+        let records: Vec<_> = batches
+            .into_iter()
+            .map(|rows| {
+                let ids: Vec<i64> = rows.iter().map(|(id, _)| *id).collect();
+                let regions: Vec<&str> = rows.iter().map(|(_, region)| *region).collect();
+                Ok(RecordBatch::try_new(
+                    schema.clone(),
+                    vec![
+                        Arc::new(Int64Array::from(ids)),
+                        Arc::new(StringArray::from(regions)),
+                    ],
+                )
+                .unwrap())
+            })
+            .collect();
+        Box::new(RecordBatchIterator::new(records, schema))
+    }
+
+    /// Create an `[id, value]` table with `id` as the unenforced primary key.
+    async fn id_value_table(dir: &TempDir) -> Table {
+        let conn = connect(dir.path().to_str().unwrap())
+            .execute()
+            .await
+            .unwrap();
+        let table = conn
+            .create_table("t", id_value_reader(vec![1, 2, 3]))
+            .execute()
+            .await
+            .unwrap();
+        table.set_unenforced_primary_key(["id"]).await.unwrap();
+        table
+    }
+
+    #[tokio::test]
+    async fn lsm_merge_insert_bucket() {
+        let dir = tempdir().unwrap();
+        let table = id_value_table(&dir).await;
+        // num_buckets = 1: every row routes to the single bucket.
+        table
+            .set_lsm_write_spec(LsmWriteSpec::bucket("id", 1))
+            .await
+            .unwrap();
+
+        // Empty `on` defaults to the primary key.
+        let mut builder = table.merge_insert(&[]);
+        builder
+            .when_matched_update_all(None)
+            .when_not_matched_insert_all();
+        let result = builder
+            .execute(id_value_reader(vec![3, 4, 5]))
+            .await
+            .unwrap();
+
+        // LSM path: rows go to the MemWAL, the breakdown is unknown until
+        // compaction, so only `num_rows` is populated.
+        assert_eq!(result.num_rows, 3);
+        assert_eq!(result.version, 0);
+        assert_eq!(result.num_inserted_rows, 0);
+        assert_eq!(result.num_updated_rows, 0);
+    }
+
+    #[tokio::test]
+    async fn lsm_merge_insert_unsharded() {
+        let dir = tempdir().unwrap();
+        let table = id_value_table(&dir).await;
+        table
+            .set_lsm_write_spec(LsmWriteSpec::unsharded())
+            .await
+            .unwrap();
+
+        let mut builder = table.merge_insert(&["id"]);
+        builder
+            .when_matched_update_all(None)
+            .when_not_matched_insert_all();
+        let result = builder
+            .execute(id_value_reader(vec![10, 11, 12, 13]))
+            .await
+            .unwrap();
+        assert_eq!(result.num_rows, 4);
+    }
+
+    #[tokio::test]
+    async fn lsm_merge_insert_identity() {
+        let dir = tempdir().unwrap();
+        let conn = connect(dir.path().to_str().unwrap())
+            .execute()
+            .await
+            .unwrap();
+        let table = conn
+            .create_table("t", id_region_reader(vec![(1, "us"), (2, "us")]))
+            .execute()
+            .await
+            .unwrap();
+        table.set_unenforced_primary_key(["id"]).await.unwrap();
+        table
+            .set_lsm_write_spec(LsmWriteSpec::identity("region"))
+            .await
+            .unwrap();
+
+        // All rows share one identity value, so they route to one shard.
+        let mut builder = table.merge_insert(&[]);
+        builder
+            .when_matched_update_all(None)
+            .when_not_matched_insert_all();
+        let result = builder
+            .execute(id_region_reader(vec![(3, "us"), (4, "us")]))
+            .await
+            .unwrap();
+        assert_eq!(result.num_rows, 2);
+    }
+
+    #[tokio::test]
+    async fn lsm_merge_insert_use_lsm_write_false_falls_back() {
+        let dir = tempdir().unwrap();
+        let table = id_value_table(&dir).await;
+        table
+            .set_lsm_write_spec(LsmWriteSpec::bucket("id", 1))
+            .await
+            .unwrap();
+
+        // use_lsm_write(false) opts out: the standard path runs and commits.
+        let mut builder = table.merge_insert(&["id"]);
+        builder.when_not_matched_insert_all().use_lsm_write(false);
+        let result = builder
+            .execute(id_value_reader(vec![3, 4, 5]))
+            .await
+            .unwrap();
+
+        assert_eq!(result.num_inserted_rows, 2);
+        assert_eq!(table.count_rows(None).await.unwrap(), 5);
+    }
+
+    #[tokio::test]
+    async fn lsm_merge_insert_rejects_on_not_primary_key() {
+        let dir = tempdir().unwrap();
+        let table = id_value_table(&dir).await;
+        table
+            .set_lsm_write_spec(LsmWriteSpec::bucket("id", 1))
+            .await
+            .unwrap();
+
+        let mut builder = table.merge_insert(&["value"]);
+        builder
+            .when_matched_update_all(None)
+            .when_not_matched_insert_all();
+        let err = builder.execute(id_value_reader(vec![1])).await.unwrap_err();
+        assert!(matches!(err, Error::InvalidInput { .. }), "got {err:?}");
+    }
+
+    #[tokio::test]
+    async fn lsm_merge_insert_rejects_non_upsert() {
+        let dir = tempdir().unwrap();
+        let table = id_value_table(&dir).await;
+        table
+            .set_lsm_write_spec(LsmWriteSpec::bucket("id", 1))
+            .await
+            .unwrap();
+
+        // Insert-only (no when_matched_update_all) is not the upsert shape.
+        let mut builder = table.merge_insert(&[]);
+        builder.when_not_matched_insert_all();
+        let err = builder.execute(id_value_reader(vec![4])).await.unwrap_err();
+        assert!(matches!(err, Error::InvalidInput { .. }), "got {err:?}");
+    }
+
+    #[tokio::test]
+    async fn lsm_close_writers_then_reopen() {
+        let dir = tempdir().unwrap();
+        let table = id_value_table(&dir).await;
+        table
+            .set_lsm_write_spec(LsmWriteSpec::bucket("id", 1))
+            .await
+            .unwrap();
+
+        let mut builder = table.merge_insert(&[]);
+        builder
+            .when_matched_update_all(None)
+            .when_not_matched_insert_all();
+        builder.execute(id_value_reader(vec![7, 8])).await.unwrap();
+
+        table.close_lsm_writers().await.unwrap();
+
+        // The writer reopens lazily on the next merge_insert.
+        let mut builder = table.merge_insert(&[]);
+        builder
+            .when_matched_update_all(None)
+            .when_not_matched_insert_all();
+        let result = builder.execute(id_value_reader(vec![9])).await.unwrap();
+        assert_eq!(result.num_rows, 1);
+    }
+
+    #[tokio::test]
+    async fn lsm_merge_insert_multi_batch() {
+        let dir = tempdir().unwrap();
+        let conn = connect(dir.path().to_str().unwrap())
+            .execute()
+            .await
+            .unwrap();
+        let table = conn
+            .create_table("t", id_region_reader(vec![(1, "us")]))
+            .execute()
+            .await
+            .unwrap();
+        table.set_unenforced_primary_key(["id"]).await.unwrap();
+        table
+            .set_lsm_write_spec(LsmWriteSpec::identity("region"))
+            .await
+            .unwrap();
+
+        // Multiple batches that all route to one shard are written together.
+        let mut builder = table.merge_insert(&[]);
+        builder
+            .when_matched_update_all(None)
+            .when_not_matched_insert_all();
+        let result = builder
+            .execute(id_region_multi_reader(vec![
+                vec![(2, "us"), (3, "us")],
+                vec![(4, "us")],
+            ]))
+            .await
+            .unwrap();
+        assert_eq!(result.num_rows, 3);
+
+        // Batches that route to different shards are rejected; the validation
+        // runs before any write, so no partial write is left behind.
+        let mut builder = table.merge_insert(&[]);
+        builder
+            .when_matched_update_all(None)
+            .when_not_matched_insert_all();
+        let err = builder
+            .execute(id_region_multi_reader(vec![
+                vec![(5, "us")],
+                vec![(6, "eu")],
+            ]))
+            .await
+            .unwrap_err();
+        assert!(matches!(err, Error::InvalidInput { .. }), "got {err:?}");
+    }
+
+    #[tokio::test]
+    async fn lsm_merge_insert_use_lsm_write_true_requires_spec() {
+        let dir = tempdir().unwrap();
+        // id_value_table sets a primary key but no LSM write spec.
+        let table = id_value_table(&dir).await;
+
+        let mut builder = table.merge_insert(&["id"]);
+        builder
+            .when_matched_update_all(None)
+            .when_not_matched_insert_all()
+            .use_lsm_write(true);
+        let err = builder.execute(id_value_reader(vec![4])).await.unwrap_err();
+        assert!(matches!(err, Error::InvalidInput { .. }), "got {err:?}");
+    }
+
+    #[tokio::test]
+    async fn lsm_merge_insert_rejects_second_shard() {
+        let dir = tempdir().unwrap();
+        let conn = connect(dir.path().to_str().unwrap())
+            .execute()
+            .await
+            .unwrap();
+        let table = conn
+            .create_table("t", id_region_reader(vec![(1, "us")]))
+            .execute()
+            .await
+            .unwrap();
+        table.set_unenforced_primary_key(["id"]).await.unwrap();
+        table
+            .set_lsm_write_spec(LsmWriteSpec::identity("region"))
+            .await
+            .unwrap();
+
+        // The first merge_insert opens the single writer for shard "us".
+        let mut builder = table.merge_insert(&[]);
+        builder
+            .when_matched_update_all(None)
+            .when_not_matched_insert_all();
+        builder
+            .execute(id_region_reader(vec![(2, "us")]))
+            .await
+            .unwrap();
+
+        // A merge_insert routing to a different shard is rejected.
+        let mut builder = table.merge_insert(&[]);
+        builder
+            .when_matched_update_all(None)
+            .when_not_matched_insert_all();
+        let err = builder
+            .execute(id_region_reader(vec![(3, "eu")]))
+            .await
+            .unwrap_err();
+        assert!(matches!(err, Error::InvalidInput { .. }), "got {err:?}");
+
+        // After closing the writer, a different shard can be written.
+        table.close_lsm_writers().await.unwrap();
+        let mut builder = table.merge_insert(&[]);
+        builder
+            .when_matched_update_all(None)
+            .when_not_matched_insert_all();
+        builder
+            .execute(id_region_reader(vec![(4, "eu")]))
+            .await
+            .unwrap();
+    }
+}

rust/lancedb/src/table/schema_evolution.rs

@@ -10,6 +10,7 @@
 
 use lance::dataset::{ColumnAlteration, NewColumnTransform};
 use serde::{Deserialize, Serialize};
+use std::collections::HashMap;
 
 use super::NativeTable;
 use crate::Result;
@@ -44,6 +45,52 @@ pub struct DropColumnsResult {
     pub version: u64,
 }
 
+/// A single field's metadata update, addressed by dot-path.
+///
+/// Merges into the field's existing metadata by default. Use [`Self::remove`] to
+/// delete a key, or [`Self::replace`] to swap the field's entire metadata map.
+#[derive(Debug, Clone, PartialEq, Eq, Default, Serialize)]
+pub struct FieldMetadataUpdate {
+    /// Dot-separated path to the field (e.g. `"embedding"` or `"address.zip"`).
+    pub path: String,
+    /// Keys to set (`Some`) or delete (`None`).
+    pub metadata: HashMap<String, Option<String>>,
+    /// If `true`, replace the field's entire metadata map instead of merging.
+    pub replace: bool,
+}
+
+impl FieldMetadataUpdate {
+    pub fn new(path: impl Into<String>) -> Self {
+        Self {
+            path: path.into(),
+            metadata: HashMap::new(),
+            replace: false,
+        }
+    }
+
+    pub fn set(mut self, key: impl Into<String>, value: impl Into<String>) -> Self {
+        self.metadata.insert(key.into(), Some(value.into()));
+        self
+    }
+
+    pub fn remove(mut self, key: impl Into<String>) -> Self {
+        self.metadata.insert(key.into(), None);
+        self
+    }
+
+    pub fn replace(mut self) -> Self {
+        self.replace = true;
+        self
+    }
+}
+
+#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, Default)]
+pub struct UpdateFieldMetadataResult {
+    /// The commit version associated with the operation.
+    #[serde(default)]
+    pub version: u64,
+}
+
 /// Internal implementation of the add columns logic.
 ///
 /// Adds new columns to the table using the provided transforms.
@@ -90,6 +137,32 @@ pub(crate) async fn execute_drop_columns(
     Ok(DropColumnsResult { version })
 }
 
+/// Internal implementation of the update field metadata logic.
+///
+/// Merges or replaces per-field metadata, addressing fields by dot-path.
+pub(crate) async fn execute_update_field_metadata(
+    table: &NativeTable,
+    updates: &[FieldMetadataUpdate],
+) -> Result<UpdateFieldMetadataResult> {
+    table.dataset.ensure_mutable()?;
+    let mut dataset = (*table.dataset.get().await?).clone();
+
+    let mut builder = dataset.update_field_metadata();
+    for update in updates {
+        let entries = update.metadata.iter().map(|(k, v)| (k.clone(), v.clone()));
+        builder = if update.replace {
+            builder.replace(&update.path, entries)?
+        } else {
+            builder.update(&update.path, entries)?
+        };
+    }
+    builder.await?;
+
+    let version = dataset.version().version;
+    table.dataset.update(dataset);
+    Ok(UpdateFieldMetadataResult { version })
+}
+
 #[cfg(test)]
 mod tests {
     use arrow_array::{Int32Array, StringArray, record_batch};
@@ -97,6 +170,7 @@ mod tests {
     use futures::TryStreamExt;
     use lance::dataset::ColumnAlteration;
 
+    use super::FieldMetadataUpdate;
     use crate::connect;
     use crate::query::{ExecutableQuery, QueryBase, Select};
     use crate::table::NewColumnTransform;
@@ -610,4 +684,46 @@ mod tests {
         let v4 = table.version().await.unwrap();
         assert_eq!(drop_result.version, v4);
     }
+
+    #[tokio::test]
+    async fn test_update_field_metadata() {
+        let conn = connect("memory://").execute().await.unwrap();
+        let batch = record_batch!(
+            ("id", Int32, [1, 2, 3]),
+            ("category", Utf8, ["A", "B", "C"])
+        )
+        .unwrap();
+        let table = conn
+            .create_table("test_update_field_metadata", batch)
+            .execute()
+            .await
+            .unwrap();
+
+        // Set metadata on a field.
+        table
+            .update_field_metadata(&[FieldMetadataUpdate::new("category")
+                .set("unit", "label")
+                .set("pii", "false")])
+            .await
+            .unwrap();
+        let schema = table.schema().await.unwrap();
+        let field = schema.field_with_name("category").unwrap();
+        assert_eq!(
+            field.metadata().get("unit").map(String::as_str),
+            Some("label")
+        );
+
+        // Merge: add a key, delete one, keep the rest.
+        table
+            .update_field_metadata(&[FieldMetadataUpdate::new("category")
+                .set("source", "import")
+                .remove("pii")])
+            .await
+            .unwrap();
+        let schema = table.schema().await.unwrap();
+        let md = schema.field_with_name("category").unwrap().metadata();
+        assert_eq!(md.get("unit").map(String::as_str), Some("label")); // preserved
+        assert_eq!(md.get("source").map(String::as_str), Some("import")); // added
+        assert!(!md.contains_key("pii")); // deleted
+    }
 }

rust/lancedb/src/utils/mod.rs

@@ -6,7 +6,7 @@ pub(crate) mod background_cache;
 use std::sync::Arc;
 
 use arrow_array::RecordBatch;
-use arrow_schema::{DataType, Schema, SchemaRef};
+use arrow_schema::{DataType, Field, Schema, SchemaRef};
 use datafusion_common::{DataFusionError, Result as DataFusionResult};
 use datafusion_execution::RecordBatchStream;
 use futures::{FutureExt, Stream};
@@ -152,14 +152,10 @@ pub fn validate_namespace(namespace: &[String]) -> Result<()> {
 /// Find one default column to create index or perform vector query.
 pub(crate) fn default_vector_column(schema: &Schema, dim: Option<i32>) -> Result<String> {
     // Try to find a vector column.
-    let candidates = schema
-        .fields()
-        .iter()
-        .filter_map(|field| match infer_vector_dim(field.data_type()) {
-            Ok(d) if dim.is_none() || dim == Some(d as i32) => Some(field.name()),
-            _ => None,
-        })
-        .collect::<Vec<_>>();
+    let mut candidates = Vec::new();
+    for field in schema.fields() {
+        collect_vector_columns(field, &mut Vec::new(), dim, &mut candidates);
+    }
     if candidates.is_empty() {
         Err(Error::InvalidInput {
             message: format!(
@@ -180,6 +176,57 @@ pub(crate) fn default_vector_column(schema: &Schema, dim: Option<i32>) -> Result
     }
 }
 
+fn collect_vector_columns(
+    field: &Field,
+    path: &mut Vec<String>,
+    dim: Option<i32>,
+    candidates: &mut Vec<String>,
+) {
+    path.push(field.name().clone());
+    match infer_vector_dim(field.data_type()) {
+        Ok(d) if dim.is_none() || dim == Some(d as i32) => {
+            let path_segments = path.iter().map(String::as_str).collect::<Vec<_>>();
+            candidates.push(lance_core::datatypes::format_field_path(&path_segments));
+        }
+        _ => {
+            if let DataType::Struct(fields) = field.data_type() {
+                for child in fields {
+                    collect_vector_columns(child, path, dim, candidates);
+                }
+            }
+        }
+    }
+    path.pop();
+}
+
+pub(crate) fn resolve_arrow_field_path(schema: &Schema, column: &str) -> Result<(String, Field)> {
+    lance_core::datatypes::parse_field_path(column).map_err(|e| Error::InvalidInput {
+        message: format!("Invalid field path `{}`: {}", column, e),
+    })?;
+
+    let lance_schema =
+        lance_core::datatypes::Schema::try_from(schema).map_err(|e| Error::Schema {
+            message: format!("Invalid schema: {}", e),
+        })?;
+    let field_path = lance_schema
+        .resolve_case_insensitive(column)
+        .ok_or_else(|| Error::Schema {
+            message: format!(
+                "Field path `{}` not found in schema. Available field paths: {}",
+                column,
+                lance_schema.field_paths().join(", ")
+            ),
+        })?;
+    let field = field_path.last().expect("field path should be non-empty");
+    let path_segments = field_path
+        .iter()
+        .map(|field| field.name.as_str())
+        .collect::<Vec<_>>();
+    let canonical_path = lance_core::datatypes::format_field_path(&path_segments);
+
+    Ok((canonical_path, Field::from(*field)))
+}
+
 pub fn supported_btree_data_type(dtype: &DataType) -> bool {
     dtype.is_integer()
         || dtype.is_floating()
@@ -450,6 +497,49 @@ mod tests {
             "vec"
         );
 
+        let schema_with_nested_vec_col = Schema::new(vec![
+            Field::new("id", DataType::Int16, true),
+            Field::new(
+                "image",
+                DataType::Struct(
+                    vec![Field::new(
+                        "embedding",
+                        DataType::FixedSizeList(
+                            Arc::new(Field::new("item", DataType::Float32, false)),
+                            10,
+                        ),
+                        false,
+                    )]
+                    .into(),
+                ),
+                false,
+            ),
+        ]);
+        assert_eq!(
+            default_vector_column(&schema_with_nested_vec_col, None).unwrap(),
+            "image.embedding"
+        );
+
+        let schema_with_escaped_nested_vec_col = Schema::new(vec![Field::new(
+            "image-meta",
+            DataType::Struct(
+                vec![Field::new(
+                    "embedding.v1",
+                    DataType::FixedSizeList(
+                        Arc::new(Field::new("item", DataType::Float32, false)),
+                        10,
+                    ),
+                    false,
+                )]
+                .into(),
+            ),
+            false,
+        )]);
+        assert_eq!(
+            default_vector_column(&schema_with_escaped_nested_vec_col, None).unwrap(),
+            "`image-meta`.`embedding.v1`"
+        );
+
         let multi_vec_col = Schema::new(vec![
             Field::new("id", DataType::Int16, true),
             Field::new(
@@ -469,6 +559,48 @@ mod tests {
                 .to_string()
                 .contains("More than one")
         );
+
+        let multi_nested_vec_col = Schema::new(vec![
+            Field::new(
+                "image",
+                DataType::Struct(
+                    vec![Field::new(
+                        "embedding",
+                        DataType::FixedSizeList(
+                            Arc::new(Field::new("item", DataType::Float32, false)),
+                            10,
+                        ),
+                        false,
+                    )]
+                    .into(),
+                ),
+                false,
+            ),
+            Field::new(
+                "text",
+                DataType::Struct(
+                    vec![Field::new(
+                        "embedding",
+                        DataType::FixedSizeList(
+                            Arc::new(Field::new("item", DataType::Float32, false)),
+                            50,
+                        ),
+                        false,
+                    )]
+                    .into(),
+                ),
+                false,
+            ),
+        ]);
+        assert_eq!(
+            default_vector_column(&multi_nested_vec_col, Some(50)).unwrap(),
+            "text.embedding"
+        );
+        let err = default_vector_column(&multi_nested_vec_col, None)
+            .unwrap_err()
+            .to_string();
+        assert!(err.contains("image.embedding"));
+        assert!(err.contains("text.embedding"));
     }
 
     #[test]