chore: update lance dependency to v7.0.0-beta.5

## Summary
- Update Lance Rust dependencies to `v7.0.0-beta.4` using
`ci/set_lance_version.py`.
- Update the Java `lance-core` dependency property to `7.0.0-beta.4`.
- Align LanceDB with dependency updates required by Lance 7, including
`object_store` 0.13 API compatibility.

Triggering tag:
https://github.com/lance-format/lance/releases/tag/v7.0.0-beta.4

## Verification
- `cargo clippy --workspace --tests --all-features -- -D warnings`
- `cargo fmt --all`

.bumpversion.toml

@@ -1,5 +1,5 @@
 [tool.bumpversion]
-current_version = "0.28.0-beta.7"
+current_version = "0.28.0-beta.11"
 parse = """(?x)
     (?P<major>0|[1-9]\\d*)\\.
     (?P<minor>0|[1-9]\\d*)\\.

.github/dependabot.yml

@@ -0,0 +1,18 @@
+version: 2
+
+# Scope: the root Cargo workspace, which produces the Rust binaries we
+# ship to users (the Node.js and Python native extensions). The
+# `rust/lancedb` library crate shares the same lockfile; its consumers
+# pick their own dependency versions, but bumping transitive deps here
+# keeps the binaries we ship current.
+updates:
+  - package-ecosystem: cargo
+    directory: /
+    schedule:
+      interval: weekly
+    open-pull-requests-limit: 10
+    groups:
+      rust-minor-patch:
+        update-types:
+          - minor
+          - patch

.github/workflows/dev.yml

@@ -8,6 +8,9 @@ concurrency:
   group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
   cancel-in-progress: true
 
+permissions:
+  contents: read
+
 jobs:
   labeler:
     permissions:

.github/workflows/java-publish.yml

@@ -19,6 +19,9 @@ on:
     paths:
       - .github/workflows/java-publish.yml
 
+permissions:
+  contents: read
+
 jobs:
   publish:
     name: Build and Publish

.github/workflows/java.yml

@@ -24,6 +24,9 @@ on:
       - java/**
       - .github/workflows/java.yml
 
+permissions:
+  contents: read
+
 jobs:
   build-java:
     runs-on: ubuntu-24.04

.github/workflows/license-header-check.yml

@@ -10,6 +10,10 @@ on:
       - nodejs/**
       - java/**
       - .github/workflows/license-header-check.yml
+
+permissions:
+  contents: read
+
 jobs:
   check-licenses:
     runs-on: ubuntu-latest

.github/workflows/nodejs.yml

@@ -15,6 +15,9 @@ on:
       - .github/workflows/nodejs.yml
       - docker-compose.yml
 
+permissions:
+  contents: read
+
 concurrency:
   group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
   cancel-in-progress: true

.github/workflows/pypi-publish.yml

@@ -14,10 +14,16 @@ on:
 env:
   PIP_EXTRA_INDEX_URL: "https://pypi.fury.io/lance-format/ https://pypi.fury.io/lancedb/"
 
+permissions:
+  contents: read
+
 jobs:
   linux:
     name: Python ${{ matrix.config.platform }} manylinux${{ matrix.config.manylinux }}
     timeout-minutes: 60
+    permissions:
+      id-token: write
+      contents: read
     strategy:
       matrix:
         config:
@@ -57,10 +63,12 @@ jobs:
       - uses: ./.github/workflows/upload_wheel
         if: startsWith(github.ref, 'refs/tags/python-v')
         with:
-          pypi_token: ${{ secrets.LANCEDB_PYPI_API_TOKEN }}
           fury_token: ${{ secrets.FURY_TOKEN }}
   mac:
     timeout-minutes: 90
+    permissions:
+      id-token: write
+      contents: read
     runs-on: ${{ matrix.config.runner }}
     strategy:
       matrix:
@@ -85,10 +93,12 @@ jobs:
       - uses: ./.github/workflows/upload_wheel
         if: startsWith(github.ref, 'refs/tags/python-v')
         with:
-          pypi_token: ${{ secrets.LANCEDB_PYPI_API_TOKEN }}
           fury_token: ${{ secrets.FURY_TOKEN }}
   windows:
     timeout-minutes: 60
+    permissions:
+      id-token: write
+      contents: read
     runs-on: windows-latest
     steps:
       - uses: actions/checkout@v4
@@ -107,7 +117,6 @@ jobs:
       - uses: ./.github/workflows/upload_wheel
         if: startsWith(github.ref, 'refs/tags/python-v')
         with:
-          pypi_token: ${{ secrets.LANCEDB_PYPI_API_TOKEN }}
           fury_token: ${{ secrets.FURY_TOKEN }}
   gh-release:
     if: startsWith(github.ref, 'refs/tags/python-v')

.github/workflows/python.yml

@@ -17,6 +17,9 @@ on:
       - .github/workflows/build_windows_wheel/**
       - .github/workflows/run_tests/**
 
+permissions:
+  contents: read
+
 concurrency:
   group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
   cancel-in-progress: true
@@ -108,7 +111,6 @@ jobs:
       - name: Install
         run: |
           pip install --extra-index-url https://pypi.fury.io/lance-format/ --extra-index-url https://pypi.fury.io/lancedb/ -e .[tests,dev,embeddings]
-          pip install tantivy
           pip install mlx
       - name: Doctest
         run: pytest --doctest-modules python/lancedb
@@ -227,6 +229,5 @@ jobs:
           pip install "pydantic<2"
           pip install pyarrow==16
           pip install --extra-index-url https://pypi.fury.io/lance-format/ --extra-index-url https://pypi.fury.io/lancedb/ -e .[tests]
-          pip install tantivy
       - name: Run tests
         run: pytest -m "not slow and not s3_test" -x -v --durations=30 python/tests

.github/workflows/rust.yml

@@ -9,9 +9,15 @@ on:
       - Cargo.toml
       - Cargo.lock
       - rust-toolchain.toml
+      - deny.toml
       - rust/**
+      - nodejs/Cargo.toml
+      - python/Cargo.toml
       - .github/workflows/rust.yml
 
+permissions:
+  contents: read
+
 concurrency:
   group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
   cancel-in-progress: true
@@ -53,6 +59,17 @@ jobs:
       - name: Run clippy (without remote feature)
         run: cargo clippy --profile ci --workspace --tests -- -D warnings
 
+  deny:
+    # Supply-chain checks: advisories, licenses, banned crates, and source
+    # restrictions. Configuration lives in `deny.toml` at the workspace root.
+    timeout-minutes: 10
+    runs-on: ubuntu-24.04
+    steps:
+      - uses: actions/checkout@v4
+      - uses: EmbarkStudios/cargo-deny-action@v2
+        with:
+          command: check advisories bans licenses sources
+
   build-no-lock:
     runs-on: ubuntu-24.04
     timeout-minutes: 30

.github/workflows/update_package_lock_run.yml

@@ -3,6 +3,9 @@ name: Update package-lock.json
 on:
   workflow_dispatch:
 
+permissions:
+  contents: read
+
 jobs:
   publish:
     runs-on: ubuntu-latest

.github/workflows/update_package_lock_run_nodejs.yml

@@ -3,6 +3,9 @@ name: Update NodeJs package-lock.json
 on:
   workflow_dispatch:
 
+permissions:
+  contents: read
+
 jobs:
   publish:
     runs-on: ubuntu-latest

.github/workflows/upload_wheel/action.yml

@@ -2,9 +2,6 @@ name: upload-wheel
 
 description: "Upload wheels to Pypi"
 inputs:
-  pypi_token:
-    required: true
-    description: "release token for the repo"
   fury_token:
     required: true
     description: "release token for the fury repo"
@@ -12,12 +9,6 @@ inputs:
 runs:
   using: "composite"
   steps:
-  - name: Install dependencies
-    shell: bash
-    run: |
-      python -m pip install --upgrade pip
-      pip install twine
-      python3 -m pip install --upgrade pkginfo
   - name: Choose repo
     shell: bash
     id: choose_repo
@@ -27,19 +18,17 @@ runs:
       else
         echo "repo=pypi" >> $GITHUB_OUTPUT
       fi
-  - name: Publish to PyPI
+  - name: Publish to Fury
+    if: steps.choose_repo.outputs.repo == 'fury'
     shell: bash
     env:
       FURY_TOKEN: ${{ inputs.fury_token }}
-      PYPI_TOKEN: ${{ inputs.pypi_token }}
     run: |
-      if [[ ${{ steps.choose_repo.outputs.repo }} == fury ]]; then
+      WHEEL=$(ls target/wheels/lancedb-*.whl 2> /dev/null | head -n 1)
-        WHEEL=$(ls target/wheels/lancedb-*.whl 2> /dev/null | head -n 1)
+      echo "Uploading $WHEEL to Fury"
-        echo "Uploading $WHEEL to Fury"
+      curl -f -F package=@$WHEEL https://$FURY_TOKEN@push.fury.io/lancedb/
-        curl -f -F package=@$WHEEL https://$FURY_TOKEN@push.fury.io/lancedb/
+  - name: Publish to PyPI
-      else
+    if: steps.choose_repo.outputs.repo == 'pypi'
-        twine upload --repository ${{ steps.choose_repo.outputs.repo }} \
+    uses: pypa/gh-action-pypi-publish@release/v1
-          --username __token__ \
+    with:
-          --password $PYPI_TOKEN \
+      packages-dir: target/wheels/
-          target/wheels/lancedb-*.whl
-      fi

Cargo.toml

@@ -1,7 +1,5 @@
 [workspace]
 members = ["rust/lancedb", "nodejs", "python"]
-# Python package needs to be built by maturin.
-exclude = ["python"]
 resolver = "2"
 
 [workspace.package]
@@ -15,40 +13,40 @@ categories = ["database-implementations"]
 rust-version = "1.91.0"
 
 [workspace.dependencies]
-lance = { "version" = "=6.0.0-beta.1", default-features = false, "tag" = "v6.0.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance = { "version" = "=7.0.0-beta.5", default-features = false, "tag" = "v7.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
-lance-core = { "version" = "=6.0.0-beta.1", "tag" = "v6.0.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-core = { "version" = "=7.0.0-beta.5", "tag" = "v7.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
-lance-datagen = { "version" = "=6.0.0-beta.1", "tag" = "v6.0.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-datagen = { "version" = "=7.0.0-beta.5", "tag" = "v7.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
-lance-file = { "version" = "=6.0.0-beta.1", "tag" = "v6.0.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-file = { "version" = "=7.0.0-beta.5", "tag" = "v7.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
-lance-io = { "version" = "=6.0.0-beta.1", default-features = false, "tag" = "v6.0.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-io = { "version" = "=7.0.0-beta.5", default-features = false, "tag" = "v7.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
-lance-index = { "version" = "=6.0.0-beta.1", "tag" = "v6.0.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-index = { "version" = "=7.0.0-beta.5", "tag" = "v7.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
-lance-linalg = { "version" = "=6.0.0-beta.1", "tag" = "v6.0.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-linalg = { "version" = "=7.0.0-beta.5", "tag" = "v7.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
-lance-namespace = { "version" = "=6.0.0-beta.1", "tag" = "v6.0.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-namespace = { "version" = "=7.0.0-beta.5", "tag" = "v7.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
-lance-namespace-impls = { "version" = "=6.0.0-beta.1", default-features = false, "tag" = "v6.0.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-namespace-impls = { "version" = "=7.0.0-beta.5", default-features = false, "tag" = "v7.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
-lance-table = { "version" = "=6.0.0-beta.1", "tag" = "v6.0.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-table = { "version" = "=7.0.0-beta.5", "tag" = "v7.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
-lance-testing = { "version" = "=6.0.0-beta.1", "tag" = "v6.0.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-testing = { "version" = "=7.0.0-beta.5", "tag" = "v7.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
-lance-datafusion = { "version" = "=6.0.0-beta.1", "tag" = "v6.0.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-datafusion = { "version" = "=7.0.0-beta.5", "tag" = "v7.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
-lance-encoding = { "version" = "=6.0.0-beta.1", "tag" = "v6.0.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-encoding = { "version" = "=7.0.0-beta.5", "tag" = "v7.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
-lance-arrow = { "version" = "=6.0.0-beta.1", "tag" = "v6.0.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-arrow = { "version" = "=7.0.0-beta.5", "tag" = "v7.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
 ahash = "0.8"
 # Note that this one does not include pyarrow
-arrow = { version = "57.2", optional = false }
+arrow = { version = "58.0.0", optional = false }
-arrow-array = "57.2"
+arrow-array = "58.0.0"
-arrow-data = "57.2"
+arrow-data = "58.0.0"
-arrow-ipc = "57.2"
+arrow-ipc = "58.0.0"
-arrow-ord = "57.2"
+arrow-ord = "58.0.0"
-arrow-schema = "57.2"
+arrow-schema = "58.0.0"
-arrow-select = "57.2"
+arrow-select = "58.0.0"
-arrow-cast = "57.2"
+arrow-cast = "58.0.0"
 async-trait = "0"
-datafusion = { version = "52.1", default-features = false }
+datafusion = { version = "53.0.0", default-features = false }
-datafusion-catalog = "52.1"
+datafusion-catalog = "53.0.0"
-datafusion-common = { version = "52.1", default-features = false }
+datafusion-common = { version = "53.0.0", default-features = false }
-datafusion-execution = "52.1"
+datafusion-execution = "53.0.0"
-datafusion-expr = "52.1"
+datafusion-expr = "53.0.0"
-datafusion-functions = "52.1"
+datafusion-functions = "53.0.0"
-datafusion-physical-plan = "52.1"
+datafusion-physical-plan = "53.0.0"
-datafusion-physical-expr = "52.1"
+datafusion-physical-expr = "53.0.0"
-datafusion-sql = "52.1"
+datafusion-sql = "53.0.0"
 env_logger = "0.11"
 half = { "version" = "2.7.1", default-features = false, features = [
     "num-traits",
@@ -56,7 +54,7 @@ half = { "version" = "2.7.1", default-features = false, features = [
 futures = "0"
 log = "0.4"
 moka = { version = "0.12", features = ["future"] }
-object_store = "0.12.0"
+object_store = "0.13.2"
 pin-project = "1.0.7"
 rand = "0.9"
 snafu = "0.8"

deny.toml

@@ -0,0 +1,172 @@
+# cargo-deny configuration for LanceDB.
+#
+# Run locally with `cargo deny check`. See
+# https://embarkstudios.github.io/cargo-deny/ for the full reference.
+
+# The set of target triples we care about. cargo-deny will only consider
+# dependencies that are used on at least one of these targets. Keeping this
+# explicit avoids noise from platform-specific crates (e.g. wasm, android,
+# ios) that we never actually ship.
+[graph]
+targets = [
+    "x86_64-unknown-linux-gnu",
+    "aarch64-unknown-linux-gnu",
+    "x86_64-apple-darwin",
+    "aarch64-apple-darwin",
+    "x86_64-pc-windows-msvc",
+    "aarch64-pc-windows-msvc",
+]
+all-features = true
+
+[output]
+feature-depth = 1
+
+# ---------------------------------------------------------------------------
+# Advisories: security vulnerabilities and yanked crates.
+# ---------------------------------------------------------------------------
+[advisories]
+version = 2
+# Fail the check if any crate in the lockfile has been yanked from crates.io.
+# Yanked crates are a signal the author retracted the release (often due to
+# bugs or security issues) and should not be depended on.
+yanked = "deny"
+# Advisory IDs we have explicitly reviewed and chosen to accept. Every
+# entry must include a rationale and, where possible, an upstream issue
+# pointing to a fix. Revisit this list whenever dependencies are updated.
+ignore = [
+    # rsa: Marvin Attack timing side-channel in PKCS#1 v1.5 decryption.
+    # Reached only through opendal → reqsign → rsa. We do not use RSA
+    # decryption in LanceDB ourselves; this is dormant in the signing path.
+    # No fixed release exists upstream as of this writing.
+    # https://rustsec.org/advisories/RUSTSEC-2023-0071
+    { id = "RUSTSEC-2023-0071", reason = "rsa crate via opendal/reqsign; no fixed upstream release" },
+
+    # instant: unmaintained. Pulled in via backoff → instant. Upstream
+    # recommends switching to `web-time`; fix has to come from backoff.
+    # https://rustsec.org/advisories/RUSTSEC-2024-0384
+    { id = "RUSTSEC-2024-0384", reason = "transitive via backoff; waiting on backoff replacement" },
+
+    # paste: unmaintained (author archived the repo). Used transitively by
+    # datafusion and the arrow ecosystem; widespread, no drop-in replacement.
+    # https://rustsec.org/advisories/RUSTSEC-2024-0436
+    { id = "RUSTSEC-2024-0436", reason = "transitive via datafusion; awaiting ecosystem migration" },
+
+    # tantivy: segfault on malformed input due to missing bounds check.
+    # Pulled in via lance for full-text search. We only feed tantivy
+    # documents we construct ourselves, not attacker-controlled bytes.
+    # Tracked for a lance dependency bump.
+    # https://rustsec.org/advisories/RUSTSEC-2025-0003
+    { id = "RUSTSEC-2025-0003", reason = "tantivy via lance; inputs are internally produced, not user-supplied bytes" },
+
+    # backoff: unmaintained. Reached only via async-openai. Replacement
+    # requires async-openai to migrate (or us to drop async-openai).
+    # https://rustsec.org/advisories/RUSTSEC-2025-0012
+    { id = "RUSTSEC-2025-0012", reason = "transitive via async-openai; waiting on upstream migration" },
+
+    # number_prefix: unmaintained. Transitive via indicatif → hf-hub.
+    # No security impact, just maintenance status.
+    # https://rustsec.org/advisories/RUSTSEC-2025-0119
+    { id = "RUSTSEC-2025-0119", reason = "transitive via hf-hub/indicatif; cosmetic formatting crate" },
+
+    # rustls-pemfile: unmaintained. Reached from two separate chains:
+    # rustls-native-certs 0.6 (via hyper-rustls 0.24) and object_store 0.12.
+    # Both upstream dependencies need to move before we can drop it.
+    # https://rustsec.org/advisories/RUSTSEC-2025-0134
+    { id = "RUSTSEC-2025-0134", reason = "transitive via rustls-native-certs/object_store; waiting on upstream migration" },
+
+    # rustls-webpki 0.101.7 (old major line): name-constraint checks for
+    # URI / wildcard names. Pulled in only via the legacy rustls 0.21 chain
+    # from aws-smithy-http-client. The 0.103 line we actively use is patched.
+    # Clearing the 0.101 copy requires the aws-sdk chain to migrate off
+    # rustls 0.21.
+    # https://rustsec.org/advisories/RUSTSEC-2026-0098
+    # https://rustsec.org/advisories/RUSTSEC-2026-0099
+    { id = "RUSTSEC-2026-0098", reason = "only affects rustls-webpki 0.101 from legacy aws-smithy/rustls 0.21 chain" },
+    { id = "RUSTSEC-2026-0099", reason = "only affects rustls-webpki 0.101 from legacy aws-smithy/rustls 0.21 chain" },
+
+    # rustls-webpki 0.101.7: reachable panic in CRL parsing. Same legacy
+    # rustls 0.21 chain from aws-smithy-http-client as above. The 0.103 line
+    # we actively use is upgraded to 0.103.13 which contains the fix.
+    # https://rustsec.org/advisories/RUSTSEC-2026-0104
+    { id = "RUSTSEC-2026-0104", reason = "only affects rustls-webpki 0.101 from legacy aws-smithy/rustls 0.21 chain" },
+]
+
+# ---------------------------------------------------------------------------
+# Licenses: only allow licenses we've reviewed as compatible with Apache-2.0.
+# ---------------------------------------------------------------------------
+[licenses]
+version = 2
+# SPDX identifiers for licenses that are compatible with our Apache-2.0
+# distribution. Additions require legal review.
+allow = [
+    "Apache-2.0",
+    "Apache-2.0 WITH LLVM-exception",
+    "MIT",
+    "BSD-2-Clause",
+    "BSD-3-Clause",
+    "ISC",
+    "Unicode-3.0",
+    "Unicode-DFS-2016",
+    "Zlib",
+    "CC0-1.0",
+    "MPL-2.0",
+    "BSL-1.0",
+    "OpenSSL",
+    # 0BSD ("BSD Zero Clause") is effectively public domain — no attribution
+    # required. Pulled in by `mock_instant`.
+    "0BSD",
+    # bzip2-1.0.6 is the permissive upstream bzip2 license (BSD-like). Pulled
+    # in by `libbz2-rs-sys`, the pure-Rust bzip2 implementation.
+    "bzip2-1.0.6",
+    # CDLA-Permissive-2.0 is a permissive data license used by `webpki-roots`
+    # for the Mozilla CA root bundle. Data-only, distribution-compatible.
+    "CDLA-Permissive-2.0",
+]
+confidence-threshold = 0.8
+# Crates whose license cannot be determined from Cargo metadata but whose
+# license we've manually confirmed from upstream. Keep this list minimal.
+[[licenses.clarify]]
+# polars-arrow-format omits the `license` field in its Cargo.toml, but the
+# upstream repo (pola-rs/polars-arrow-format) is dual-licensed Apache-2.0 OR
+# MIT. See https://github.com/pola-rs/polars-arrow-format/blob/main/LICENSE
+crate = "polars-arrow-format"
+expression = "Apache-2.0 OR MIT"
+license-files = []
+
+# ---------------------------------------------------------------------------
+# Bans: disallow specific crates and flag dependency hygiene issues.
+# ---------------------------------------------------------------------------
+[bans]
+# Warn (not deny) on duplicate versions of the same crate. In a large
+# workspace like this one, duplicates are common and often unavoidable
+# transitively. We surface them to discourage growth, but don't fail CI.
+multiple-versions = "warn"
+# Wildcard version requirements (`foo = "*"`) are a footgun — they let any
+# future release in without review. Ban them outright.
+wildcards = "deny"
+# Internal workspace crates reference each other via `path = "..."`, which
+# cargo-deny sees as a wildcard version. That's fine for private workspace
+# members (not published to crates.io), so allow it specifically for paths.
+allow-wildcard-paths = true
+# Features that, if enabled, should cause the check to fail.
+deny = []
+# Crates to skip when checking for duplicate versions.
+skip = []
+# Similar to `skip`, but also skips the entire transitive subtree.
+skip-tree = []
+
+# ---------------------------------------------------------------------------
+# Sources: restrict where crates can come from.
+# ---------------------------------------------------------------------------
+[sources]
+# Deny any registry other than the ones explicitly listed below.
+unknown-registry = "deny"
+# Deny any git dependency whose host isn't in the allow-list below. This
+# prevents accidental pulls from arbitrary forks.
+unknown-git = "deny"
+allow-registry = ["https://github.com/rust-lang/crates.io-index"]
+# Lance is developed in a sibling repo and pulled as a git dependency until
+# releases are cut to crates.io. Allow that specific host.
+allow-git = [
+    "https://github.com/lance-format/lance",
+]

dockerfiles/Dockerfile

@@ -24,4 +24,4 @@ RUN python --version && \
   rustc --version && \
   protoc --version
 
-RUN pip install --no-cache-dir tantivy lancedb
+RUN pip install --no-cache-dir lancedb

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>0.28.0-beta.7</version>
+    <version>0.28.0-beta.11</version>
 </dependency>
 ```
 

docs/src/js/interfaces/ConnectionOptions.md

@@ -41,6 +41,29 @@ for testing purposes.
 
 ***
 
+### manifestEnabled?
+
+```ts
+optional manifestEnabled: boolean;
+```
+
+(For LanceDB OSS only): use directory namespace manifests as the source
+of truth for table metadata. Existing directory-listed root tables are
+migrated into the manifest on access.
+
+***
+
+### namespaceClientProperties?
+
+```ts
+optional namespaceClientProperties: Record<string, string>;
+```
+
+(For LanceDB OSS only): extra properties for the backing namespace
+client used by manifest-enabled native connections.
+
+***
+
 ### readConsistencyInterval?
 
 ```ts

java/lancedb-core/pom.xml

@@ -8,7 +8,7 @@
     <parent>
       <groupId>com.lancedb</groupId>
       <artifactId>lancedb-parent</artifactId>
-      <version>0.28.0-beta.7</version>
+      <version>0.28.0-beta.11</version>
       <relativePath>../pom.xml</relativePath>
     </parent>
 

java/pom.xml

@@ -6,7 +6,7 @@
 
     <groupId>com.lancedb</groupId>
     <artifactId>lancedb-parent</artifactId>
-    <version>0.28.0-beta.7</version>
+    <version>0.28.0-beta.11</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>6.0.0-beta.1</lance-core.version>
+        <lance-core.version>7.0.0-beta.5</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 +1,8 @@
 [package]
 name = "lancedb-nodejs"
 edition.workspace = true
-version = "0.28.0-beta.7"
+version = "0.28.0-beta.11"
+publish = false
 license.workspace = true
 description.workspace = true
 repository.workspace = true
@@ -15,7 +16,7 @@ crate-type = ["cdylib"]
 async-trait.workspace = true
 arrow-ipc.workspace = true
 arrow-array.workspace = true
-arrow-buffer = "57.2"
+arrow-buffer = "58.0.0"
 half.workspace = true
 arrow-schema.workspace = true
 env_logger.workspace = true
@@ -31,8 +32,8 @@ lzma-sys = { version = "0.1", features = ["static"] }
 log.workspace = true
 
 # Pin to resolve build failures; update periodically for security patches.
-aws-lc-sys = "=0.38.0"
+aws-lc-sys = "=0.40.0"
-aws-lc-rs = "=1.16.1"
+aws-lc-rs = "=1.16.3"
 
 [build-dependencies]
 napi-build = "2.3.1"

nodejs/__test__/rerankers.test.ts

@@ -1,6 +1,8 @@
 // SPDX-License-Identifier: Apache-2.0
 // SPDX-FileCopyrightText: Copyright The LanceDB Authors
 
+import { spawn } from "node:child_process";
+import * as path from "node:path";
 import { RecordBatch } from "apache-arrow";
 import * as tmp from "tmp";
 import { Connection, Index, Table, connect, makeArrowTable } from "../lancedb";
@@ -76,4 +78,91 @@ describe("rerankers", function () {
 
     expect(result).toHaveLength(2);
   });
+
+  it("does not keep process alive after rerank query", async function () {
+    const script = `
+import * as lancedb from "./dist/index.js";
+import * as os from "node:os";
+import * as path from "node:path";
+import * as fs from "node:fs/promises";
+
+const dir = await fs.mkdtemp(path.join(os.tmpdir(), "lancedb-rerank-exit-"));
+const db = await lancedb.connect(dir);
+const table = await db.createTable("test", [{ text: "hello", vector: [1, 2, 3] }], {
+  mode: "overwrite",
+});
+await table.createIndex("text", { config: lancedb.Index.fts() });
+await table.waitForIndex(["text_idx"], 30);
+
+const reranker = await lancedb.rerankers.RRFReranker.create();
+await table
+  .query()
+  .nearestTo([1, 2, 3])
+  .fullTextSearch("hello")
+  .rerank(reranker)
+  .toArray();
+
+table.close();
+db.close();
+`;
+
+    await new Promise<void>((resolve, reject) => {
+      const child = spawn(
+        process.execPath,
+        ["--input-type=module", "-e", script],
+        {
+          cwd: path.resolve(__dirname, ".."),
+          stdio: ["ignore", "pipe", "pipe"],
+        },
+      );
+
+      let stdout = "";
+      let stderr = "";
+
+      child.stdout.on("data", (chunk) => {
+        stdout += chunk.toString();
+      });
+
+      child.stderr.on("data", (chunk) => {
+        stderr += chunk.toString();
+      });
+
+      const timeout = setTimeout(() => {
+        child.kill();
+        reject(
+          new Error(
+            `child process did not exit in time\nstdout:\n${stdout}\nstderr:\n${stderr}`,
+          ),
+        );
+      }, 20_000);
+
+      child.on("error", (err) => {
+        clearTimeout(timeout);
+        reject(err);
+      });
+
+      child.on("exit", (code, signal) => {
+        clearTimeout(timeout);
+        if (signal !== null) {
+          reject(
+            new Error(
+              `child process exited with signal ${signal}\nstdout:\n${stdout}\nstderr:\n${stderr}`,
+            ),
+          );
+          return;
+        }
+
+        if (code !== 0) {
+          reject(
+            new Error(
+              `child process exited with code ${code}\nstdout:\n${stdout}\nstderr:\n${stderr}`,
+            ),
+          );
+          return;
+        }
+
+        resolve();
+      });
+    });
+  });
 });

nodejs/npm/darwin-arm64/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-darwin-arm64",
-	"version": "0.28.0-beta.7",
+	"version": "0.28.0-beta.11",
 	"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": "0.28.0-beta.7",
+	"version": "0.28.0-beta.11",
 	"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": "0.28.0-beta.7",
+	"version": "0.28.0-beta.11",
 	"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": "0.28.0-beta.7",
+	"version": "0.28.0-beta.11",
 	"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": "0.28.0-beta.7",
+	"version": "0.28.0-beta.11",
 	"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": "0.28.0-beta.7",
+  "version": "0.28.0-beta.11",
   "os": [
     "win32"
   ],

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

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-win32-x64-msvc",
-	"version": "0.28.0-beta.7",
+	"version": "0.28.0-beta.11",
 	"os": ["win32"],
 	"cpu": ["x64"],
 	"main": "lancedb.win32-x64-msvc.node",

nodejs/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "@lancedb/lancedb",
-  "version": "0.28.0-beta.7",
+  "version": "0.28.0-beta.11",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "@lancedb/lancedb",
-      "version": "0.28.0-beta.7",
+      "version": "0.28.0-beta.11",
       "cpu": [
         "x64",
         "arm64"

nodejs/package.json

@@ -11,7 +11,7 @@
     "ann"
   ],
   "private": false,
-  "version": "0.28.0-beta.7",
+  "version": "0.28.0-beta.11",
   "main": "dist/index.js",
   "exports": {
     ".": "./dist/index.js",
@@ -75,7 +75,6 @@
     "build:debug": "napi build --platform --dts ../lancedb/native.d.ts --js ../lancedb/native.js --output-dir lancedb",
     "postbuild:debug": "shx mkdir -p dist && shx cp lancedb/*.node dist/",
     "build:release": "napi build --platform --release --dts ../lancedb/native.d.ts --js ../lancedb/native.js --output-dir dist",
-    "postbuild:release": "shx mkdir -p dist && shx cp lancedb/*.node dist/",
     "build": "npm run build:debug && npm run tsc",
     "build-release": "npm run build:release && npm run tsc",
     "tsc": "tsc -b",

nodejs/src/connection.rs

@@ -67,6 +67,12 @@ impl Connection {
                 builder = builder.storage_option(key, value);
             }
         }
+        if let Some(manifest_enabled) = options.manifest_enabled {
+            builder = builder.manifest_enabled(manifest_enabled);
+        }
+        if let Some(namespace_client_properties) = options.namespace_client_properties {
+            builder = builder.namespace_client_properties(namespace_client_properties);
+        }
 
         // Create client config, optionally with header provider
         let client_config = options.client_config.unwrap_or_default();

nodejs/src/lib.rs

@@ -37,6 +37,13 @@ pub struct ConnectionOptions {
     ///
     /// The available options are described at https://docs.lancedb.com/storage/
     pub storage_options: Option<HashMap<String, String>>,
+    /// (For LanceDB OSS only): use directory namespace manifests as the source
+    /// of truth for table metadata. Existing directory-listed root tables are
+    /// migrated into the manifest on access.
+    pub manifest_enabled: Option<bool>,
+    /// (For LanceDB OSS only): extra properties for the backing namespace
+    /// client used by manifest-enabled native connections.
+    pub namespace_client_properties: Option<HashMap<String, String>>,
     /// (For LanceDB OSS only): the session to use for this connection. Holds
     /// shared caches and other session-specific state.
     pub session: Option<session::Session>,

nodejs/src/rerankers.rs

@@ -18,6 +18,7 @@ type RerankHybridFn = ThreadsafeFunction<
     RerankHybridCallbackArgs,
     Status,
     false,
+    true,
 >;
 
 /// Reranker implementation that "wraps" a NodeJS Reranker implementation.
@@ -32,7 +33,10 @@ impl Reranker {
     pub fn new(
         rerank_hybrid: Function<RerankHybridCallbackArgs, Promise<Buffer>>,
     ) -> napi::Result<Self> {
-        let rerank_hybrid = rerank_hybrid.build_threadsafe_function().build()?;
+        let rerank_hybrid = rerank_hybrid
+            .build_threadsafe_function()
+            .weak::<true>()
+            .build()?;
         Ok(Self { rerank_hybrid })
     }
 }

python/.bumpversion.toml

@@ -1,5 +1,5 @@
 [tool.bumpversion]
-current_version = "0.31.0-beta.7"
+current_version = "0.31.0-beta.11"
 parse = """(?x)
     (?P<major>0|[1-9]\\d*)\\.
     (?P<minor>0|[1-9]\\d*)\\.

python/Cargo.toml

@@ -1,6 +1,7 @@
 [package]
 name = "lancedb-python"
-version = "0.31.0-beta.7"
+version = "0.31.0-beta.11"
+publish = false
 edition.workspace = true
 description = "Python bindings for LanceDB"
 license.workspace = true
@@ -14,7 +15,7 @@ name = "_lancedb"
 crate-type = ["cdylib"]
 
 [dependencies]
-arrow = { version = "57.2", features = ["pyarrow"] }
+arrow = { version = "58.0.0", features = ["pyarrow"] }
 async-trait = "0.1"
 bytes = "1"
 lancedb = { path = "../rust/lancedb", default-features = false }
@@ -24,8 +25,8 @@ lance-namespace-impls.workspace = true
 lance-io.workspace = true
 env_logger.workspace = true
 log.workspace = true
-pyo3 = { version = "0.26", features = ["extension-module", "abi3-py39"] }
+pyo3 = { version = "0.28", features = ["extension-module", "abi3-py39"] }
-pyo3-async-runtimes = { version = "0.26", features = [
+pyo3-async-runtimes = { version = "0.28", features = [
     "attributes",
     "tokio-runtime",
 ] }
@@ -34,10 +35,11 @@ futures.workspace = true
 serde = "1"
 serde_json = "1"
 snafu.workspace = true
-tokio = { version = "1.40", features = ["sync"] }
+tokio = { version = "1.40", features = ["sync", "rt-multi-thread"] }
+libc = "0.2"
 
 [build-dependencies]
-pyo3-build-config = { version = "0.26", features = [
+pyo3-build-config = { version = "0.28", features = [
     "extension-module",
     "abi3-py39",
 ] }

python/PYTHON_THIRD_PARTY_LICENSES.md

@@ -183,7 +183,6 @@
 | stack-data                     | 0.6.3           | MIT License                                                                                      | http://github.com/alexmojaki/stack_data                                                           |
 | sympy                          | 1.14.0          | BSD License                                                                                      | https://sympy.org                                                                                 |
 | tabulate                       | 0.9.0           | MIT License                                                                                      | https://github.com/astanin/python-tabulate                                                        |
-| tantivy                        | 0.25.1          | UNKNOWN                                                                                          | UNKNOWN                                                                                           |
 | threadpoolctl                  | 3.6.0           | BSD License                                                                                      | https://github.com/joblib/threadpoolctl                                                           |
 | timm                           | 1.0.24          | Apache Software License                                                                          | https://github.com/huggingface/pytorch-image-models                                               |
 | tinycss2                       | 1.4.0           | BSD License                                                                                      | https://www.courtbouillon.org/tinycss2                                                            |

python/pyproject.toml

@@ -57,7 +57,6 @@ tests = [
     "duckdb>=0.9.0",
     "pytz>=2023.3",
     "polars>=0.19, <=1.3.0",
-    "tantivy>=0.20.0",
     "pyarrow-stubs>=16.0",
     "pylance>=5.0.0b5",
     "requests>=2.31.0",

python/python/lancedb/__init__.py

@@ -7,7 +7,6 @@ import os
 from concurrent.futures import ThreadPoolExecutor
 from datetime import timedelta
 from typing import Dict, Optional, Union, Any, List
-import warnings
 
 __version__ = importlib.metadata.version("lancedb")
 
@@ -73,6 +72,7 @@ def connect(
     client_config: Union[ClientConfig, Dict[str, Any], None] = None,
     storage_options: Optional[Dict[str, str]] = None,
     session: Optional[Session] = None,
+    manifest_enabled: bool = False,
     namespace_client_impl: Optional[str] = None,
     namespace_client_properties: Optional[Dict[str, str]] = None,
     namespace_client_pushdown_operations: Optional[List[str]] = None,
@@ -111,6 +111,10 @@ def connect(
     storage_options: dict, optional
         Additional options for the storage backend. See available options at
         <https://docs.lancedb.com/storage/>
+    manifest_enabled : bool, default False
+        When true for local/native connections, use directory namespace
+        manifests as the source of truth for table metadata. Existing
+        directory-listed root tables are migrated into the manifest on access.
     session: Session, optional
         (For LanceDB OSS only)
         A session to use for this connection. Sessions allow you to configure
@@ -158,11 +162,11 @@ def connect(
     conn : DBConnection
         A connection to a LanceDB database.
     """
-    if namespace_client_impl is not None or namespace_client_properties is not None:
+    if namespace_client_impl is not None:
-        if namespace_client_impl is None or namespace_client_properties is None:
+        if namespace_client_properties is None:
             raise ValueError(
-                "Both namespace_client_impl and "
+                "namespace_client_properties must be provided when "
-                "namespace_client_properties must be provided"
+                "namespace_client_impl is set"
             )
         if kwargs:
             raise ValueError(f"Unknown keyword arguments: {kwargs}")
@@ -175,6 +179,12 @@ def connect(
             namespace_client_pushdown_operations=namespace_client_pushdown_operations,
         )
 
+    if namespace_client_properties is not None and not manifest_enabled:
+        raise ValueError(
+            "namespace_client_impl must be provided when using "
+            "namespace_client_properties unless manifest_enabled=True"
+        )
+
     if namespace_client_pushdown_operations is not None:
         raise ValueError(
             "namespace_client_pushdown_operations is only valid when "
@@ -212,6 +222,8 @@ def connect(
         read_consistency_interval=read_consistency_interval,
         storage_options=storage_options,
         session=session,
+        manifest_enabled=manifest_enabled,
+        namespace_client_properties=namespace_client_properties,
     )
 
 
@@ -289,6 +301,8 @@ def deserialize_conn(
             parsed["uri"],
             read_consistency_interval=rci,
             storage_options=storage_options,
+            manifest_enabled=parsed.get("manifest_enabled", False),
+            namespace_client_properties=parsed.get("namespace_client_properties"),
         )
     else:
         raise ValueError(f"Unknown connection_type: {connection_type}")
@@ -304,6 +318,8 @@ async def connect_async(
     client_config: Optional[Union[ClientConfig, Dict[str, Any]]] = None,
     storage_options: Optional[Dict[str, str]] = None,
     session: Optional[Session] = None,
+    manifest_enabled: bool = False,
+    namespace_client_properties: Optional[Dict[str, str]] = None,
 ) -> AsyncConnection:
     """Connect to a LanceDB database.
 
@@ -343,6 +359,13 @@ async def connect_async(
         cache sizes for index and metadata caches, which can significantly
         impact memory use and performance. They can also be re-used across
         multiple connections to share the same cache state.
+    manifest_enabled : bool, default False
+        When true for local/native connections, use directory namespace
+        manifests as the source of truth for table metadata. Existing
+        directory-listed root tables are migrated into the manifest on access.
+    namespace_client_properties : dict, optional
+        Additional directory namespace client properties to use with
+        ``manifest_enabled=True``.
 
     Examples
     --------
@@ -385,6 +408,8 @@ async def connect_async(
             client_config,
             storage_options,
             session,
+            manifest_enabled,
+            namespace_client_properties,
         )
     )
 
@@ -412,13 +437,3 @@ __all__ = [
     "Table",
     "__version__",
 ]
-
-
-def __warn_on_fork():
-    warnings.warn(
-        "lance is not fork-safe. If you are using multiprocessing, use spawn instead.",
-    )
-
-
-if hasattr(os, "register_at_fork"):
-    os.register_at_fork(before=__warn_on_fork)  # type: ignore[attr-defined]

python/python/lancedb/_lancedb.pyi

@@ -242,6 +242,8 @@ async def connect(
     client_config: Optional[Union[ClientConfig, Dict[str, Any]]],
     storage_options: Optional[Dict[str, str]],
     session: Optional[Session],
+    manifest_enabled: bool = False,
+    namespace_client_properties: Optional[Dict[str, str]] = None,
 ) -> Connection: ...
 
 class RecordBatchStream:
@@ -440,7 +442,7 @@ class AsyncPermutationBuilder:
     async def execute(self) -> Table: ...
 
 def async_permutation_builder(
-    table: Table, dest_table_name: str
+    table: Table,
 ) -> AsyncPermutationBuilder: ...
 def fts_query_to_json(query: Any) -> str: ...
 

python/python/lancedb/background_loop.py

@@ -2,7 +2,9 @@
 # SPDX-FileCopyrightText: Copyright The LanceDB Authors
 
 import asyncio
+import os
 import threading
+import warnings
 
 
 class BackgroundEventLoop:
@@ -13,6 +15,9 @@ class BackgroundEventLoop:
     """
 
     def __init__(self):
+        self._start()
+
+    def _start(self):
         self.loop = asyncio.new_event_loop()
         self.thread = threading.Thread(
             target=self.loop.run_forever,
@@ -31,3 +36,30 @@ class BackgroundEventLoop:
 
 
 LOOP = BackgroundEventLoop()
+
+_FORK_WARNED = False
+
+
+def _reset_after_fork():
+    # Threads do not survive fork(), so the asyncio loop in LOOP.thread is
+    # dead in the child. Re-initialize the singleton in place so existing
+    # `from .background_loop import LOOP` references in other modules see
+    # the new state. The Rust-side tokio runtime is reset analogously by a
+    # pthread_atfork hook installed in the _lancedb extension.
+    LOOP._start()
+    global _FORK_WARNED
+    if not _FORK_WARNED:
+        _FORK_WARNED = True
+        warnings.warn(
+            "lancedb fork support is experimental: the internal async "
+            "runtime has been reset in the forked child, but a small chance "
+            "of deadlock remains if other state was mid-operation at fork "
+            "time. The 'forkserver' or 'spawn' multiprocessing start method "
+            "is likely a safer alternative.",
+            RuntimeWarning,
+            stacklevel=2,
+        )
+
+
+if hasattr(os, "register_at_fork"):
+    os.register_at_fork(after_in_child=_reset_after_fork)

python/python/lancedb/db.py

@@ -590,8 +590,13 @@ class LanceDBConnection(DBConnection):
         read_consistency_interval: Optional[timedelta] = None,
         storage_options: Optional[Dict[str, str]] = None,
         session: Optional[Session] = None,
+        manifest_enabled: bool = False,
+        namespace_client_properties: Optional[Dict[str, str]] = None,
         _inner: Optional[LanceDbConnection] = None,
     ):
+        self.storage_options = storage_options
+        self._manifest_enabled = manifest_enabled
+        self._namespace_client_properties = namespace_client_properties
         if _inner is not None:
             self._conn = _inner
             self._cached_namespace_client = None
@@ -633,6 +638,8 @@ class LanceDBConnection(DBConnection):
                 None,
                 storage_options,
                 session,
+                manifest_enabled,
+                namespace_client_properties,
             )
 
         # TODO: It would be nice if we didn't store self.storage_options but it is
@@ -640,7 +647,6 @@ class LanceDBConnection(DBConnection):
         # work because some paths like LanceDBConnection.from_inner will lose the
         # storage_options.  Also, this class really shouldn't be holding any state
         # beyond _conn.
-        self.storage_options = storage_options
         self._conn = AsyncConnection(LOOP.run(do_connect()))
         self._cached_namespace_client: Optional[LanceNamespace] = None
 
@@ -677,6 +683,8 @@ class LanceDBConnection(DBConnection):
                 "connection_type": "local",
                 "uri": self.uri,
                 "storage_options": self.storage_options,
+                "manifest_enabled": self._manifest_enabled,
+                "namespace_client_properties": self._namespace_client_properties,
                 "read_consistency_interval_seconds": (
                     rci.total_seconds() if rci else None
                 ),

python/python/lancedb/fts.py

@@ -1,201 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright The LanceDB Authors
-
-"""Full text search index using tantivy-py"""
-
-import os
-from typing import List, Tuple, Optional
-
-import pyarrow as pa
-
-try:
-    import tantivy
-except ImportError:
-    raise ImportError(
-        "Please install tantivy-py `pip install tantivy` to use the full text search feature."  # noqa: E501
-    )
-
-from .table import LanceTable
-
-
-def create_index(
-    index_path: str,
-    text_fields: List[str],
-    ordering_fields: Optional[List[str]] = None,
-    tokenizer_name: str = "default",
-) -> tantivy.Index:
-    """
-    Create a new Index (not populated)
-
-    Parameters
-    ----------
-    index_path : str
-        Path to the index directory
-    text_fields : List[str]
-        List of text fields to index
-    ordering_fields: List[str]
-        List of unsigned type fields to order by at search time
-    tokenizer_name : str, default "default"
-        The tokenizer to use
-
-    Returns
-    -------
-    index : tantivy.Index
-        The index object (not yet populated)
-    """
-    if ordering_fields is None:
-        ordering_fields = []
-    # Declaring our schema.
-    schema_builder = tantivy.SchemaBuilder()
-    # special field that we'll populate with row_id
-    schema_builder.add_integer_field("doc_id", stored=True)
-    # data fields
-    for name in text_fields:
-        schema_builder.add_text_field(name, stored=True, tokenizer_name=tokenizer_name)
-    if ordering_fields:
-        for name in ordering_fields:
-            schema_builder.add_unsigned_field(name, fast=True)
-    schema = schema_builder.build()
-    os.makedirs(index_path, exist_ok=True)
-    index = tantivy.Index(schema, path=index_path)
-    return index
-
-
-def populate_index(
-    index: tantivy.Index,
-    table: LanceTable,
-    fields: List[str],
-    writer_heap_size: Optional[int] = None,
-    ordering_fields: Optional[List[str]] = None,
-) -> int:
-    """
-    Populate an index with data from a LanceTable
-
-    Parameters
-    ----------
-    index : tantivy.Index
-        The index object
-    table : LanceTable
-        The table to index
-    fields : List[str]
-        List of fields to index
-    writer_heap_size : int
-        The writer heap size in bytes, defaults to 1GB
-
-    Returns
-    -------
-    int
-        The number of rows indexed
-    """
-    if ordering_fields is None:
-        ordering_fields = []
-    writer_heap_size = writer_heap_size or 1024 * 1024 * 1024
-    # first check the fields exist and are string or large string type
-    nested = []
-
-    for name in fields:
-        try:
-            f = table.schema.field(name)  # raises KeyError if not found
-        except KeyError:
-            f = resolve_path(table.schema, name)
-            nested.append(name)
-
-        if not pa.types.is_string(f.type) and not pa.types.is_large_string(f.type):
-            raise TypeError(f"Field {name} is not a string type")
-
-    # create a tantivy writer
-    writer = index.writer(heap_size=writer_heap_size)
-    # write data into index
-    dataset = table.to_lance()
-    row_id = 0
-
-    max_nested_level = 0
-    if len(nested) > 0:
-        max_nested_level = max([len(name.split(".")) for name in nested])
-
-    for b in dataset.to_batches(columns=fields + ordering_fields):
-        if max_nested_level > 0:
-            b = pa.Table.from_batches([b])
-            for _ in range(max_nested_level - 1):
-                b = b.flatten()
-        for i in range(b.num_rows):
-            doc = tantivy.Document()
-            for name in fields:
-                value = b[name][i].as_py()
-                if value is not None:
-                    doc.add_text(name, value)
-            for name in ordering_fields:
-                value = b[name][i].as_py()
-                if value is not None:
-                    doc.add_unsigned(name, value)
-            if not doc.is_empty:
-                doc.add_integer("doc_id", row_id)
-                writer.add_document(doc)
-            row_id += 1
-    # commit changes
-    writer.commit()
-    return row_id
-
-
-def resolve_path(schema, field_name: str) -> pa.Field:
-    """
-    Resolve a nested field path to a list of field names
-
-    Parameters
-    ----------
-    field_name : str
-        The field name to resolve
-
-    Returns
-    -------
-    List[str]
-        The resolved path
-    """
-    path = field_name.split(".")
-    field = schema.field(path.pop(0))
-    for segment in path:
-        if pa.types.is_struct(field.type):
-            field = field.type.field(segment)
-        else:
-            raise KeyError(f"field {field_name} not found in schema {schema}")
-    return field
-
-
-def search_index(
-    index: tantivy.Index, query: str, limit: int = 10, ordering_field=None
-) -> Tuple[Tuple[int], Tuple[float]]:
-    """
-    Search an index for a query
-
-    Parameters
-    ----------
-    index : tantivy.Index
-        The index object
-    query : str
-        The query string
-    limit : int
-        The maximum number of results to return
-
-    Returns
-    -------
-    ids_and_score: list[tuple[int], tuple[float]]
-        A tuple of two tuples, the first containing the document ids
-        and the second containing the scores
-    """
-    searcher = index.searcher()
-    query = index.parse_query(query)
-    # get top results
-    if ordering_field:
-        results = searcher.search(query, limit, order_by_field=ordering_field)
-    else:
-        results = searcher.search(query, limit)
-    if results.count == 0:
-        return tuple(), tuple()
-    return tuple(
-        zip(
-            *[
-                (searcher.doc(doc_address)["doc_id"][0], score)
-                for score, doc_address in results.hits
-            ]
-        )
-    )

python/python/lancedb/permutation.py

@@ -1,11 +1,12 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright The LanceDB Authors
 
-from deprecation import deprecated
+import copy
-from lancedb import AsyncConnection, DBConnection
-import pyarrow as pa
 import json
 
+from deprecation import deprecated
+import pyarrow as pa
+
 from ._lancedb import async_permutation_builder, PermutationReader
 from .table import LanceTable
 from .background_loop import LOOP
@@ -36,10 +37,7 @@ class PermutationBuilder:
     be referenced by name in the future.  If names are not provided then they can only
     be referenced by their ordinal index.  There is no requirement to name every split.
 
-    By default, the permutation will be stored in memory and will be lost when the
+    The permutation is stored in memory and will be lost when the program exits.
-    program exits.  To persist the permutation (for very large datasets or to share
-    the permutation across multiple workers) use the [persist](#persist) method to
-    create a permanent table.
     """
 
     def __init__(self, table: LanceTable):
@@ -51,15 +49,6 @@ class PermutationBuilder:
         """
         self._async = async_permutation_builder(table)
 
-    def persist(
-        self, database: Union[DBConnection, AsyncConnection], table_name: str
-    ) -> "PermutationBuilder":
-        """
-        Persist the permutation to the given database.
-        """
-        self._async.persist(database, table_name)
-        return self
-
     def split_random(
         self,
         *,
@@ -380,20 +369,44 @@ class Permutation:
 
     def __init__(
         self,
-        reader: PermutationReader,
+        base_table: LanceTable,
+        permutation_table: Optional[LanceTable],
+        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,
+        _reader: Optional[PermutationReader] = None,
     ):
         """
         Internal constructor.  Use [from_tables](#from_tables) instead.
         """
-        assert reader is not None, "reader is required"
+        assert base_table is not None, "base_table is required"
         assert selection is not None, "selection is required"
-        self.reader = reader
+        self.base_table = base_table
+        self.permutation_table = permutation_table
+        self.split = split
         self.selection = selection
         self.transform_fn = transform_fn
         self.batch_size = batch_size
+        self.offset = offset
+        self.limit = limit
+        self.connection_factory = connection_factory
+        if _reader is None:
+            _reader = LOOP.run(self._build_reader())
+        self.reader: PermutationReader = _reader
+
+    async def _build_reader(self) -> PermutationReader:
+        reader = await PermutationReader.from_tables(
+            self.base_table, self.permutation_table, self.split
+        )
+        if self.offset is not None:
+            reader = await reader.with_offset(self.offset)
+        if self.limit is not None:
+            reader = await reader.with_limit(self.limit)
+        return reader
 
     def _with_selection(self, selection: dict[str, str]) -> "Permutation":
         """
@@ -402,21 +415,97 @@ class Permutation:
         Does not validation of the selection and it replaces it entirely.  This is not
         intended for public use.
         """
-        return Permutation(self.reader, selection, self.batch_size, self.transform_fn)
+        new = copy.copy(self)
-
+        new.selection = selection
-    def _with_reader(self, reader: PermutationReader) -> "Permutation":
+        return new
-        """
-        Creates a new permutation with the given reader
-
-        This is an internal method and should not be used directly.
-        """
-        return Permutation(reader, self.selection, self.batch_size, self.transform_fn)
 
     def with_batch_size(self, batch_size: int) -> "Permutation":
         """
         Creates a new permutation with the given batch size
         """
-        return Permutation(self.reader, self.selection, batch_size, self.transform_fn)
+        new = copy.copy(self)
+        new.batch_size = batch_size
+        return new
+
+    def with_connection_factory(
+        self, connection_factory: Callable[[str], LanceTable]
+    ) -> "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
+        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=...)``.
+
+        Examples
+        --------
+        Basic native (file-system path), parameterized via ``functools.partial``::
+
+            import functools, lancedb
+            from lancedb.permutation import Permutation
+
+            def open_native_table(uri: str, table_name: str):
+                return lancedb.connect(uri).open_table(table_name)
+
+            factory = functools.partial(open_native_table, "/data/lance_db")
+            permutation = Permutation.identity(
+                factory("training")
+            ).with_connection_factory(factory)
+
+        Native via :func:`lancedb.connect_namespace` (e.g. a directory- or
+        REST-backed namespace client). The factory takes the
+        implementation name and properties dict as partial-bound args so
+        the worker can rebuild the same namespace connection::
+
+            def open_via_namespace(
+                impl: str, properties: dict[str, str], table_name: str,
+            ):
+                return lancedb.connect_namespace(impl, properties).open_table(
+                    table_name,
+                )
+
+            factory = functools.partial(
+                open_via_namespace,
+                "dir",
+                {"root": "/data/lance_db"},
+            )
+
+        LanceDB Cloud, reading credentials from env vars at worker startup
+        so secrets aren't pickled into the dataset::
+
+            import os, lancedb
+
+            def open_remote_table(table_name: str):
+                db = lancedb.connect(
+                    "db://my-database",
+                    api_key=os.environ["LANCEDB_API_KEY"],
+                    region=os.environ.get("LANCEDB_REGION", "us-east-1"),
+                )
+                return db.open_table(table_name)
+
+            permutation = Permutation.identity(
+                open_remote_table("training")
+            ).with_connection_factory(open_remote_table)
+        """
+        assert connection_factory is not None, "connection_factory is required"
+        new = copy.copy(self)
+        new.connection_factory = connection_factory
+        return new
 
     @classmethod
     def identity(cls, table: LanceTable) -> "Permutation":
@@ -489,11 +578,126 @@ class Permutation:
             schema = await reader.output_schema(None)
             initial_selection = {name: name for name in schema.names}
             return cls(
-                reader, initial_selection, DEFAULT_BATCH_SIZE, Transforms.arrow2python
+                base_table,
+                permutation_table,
+                split,
+                initial_selection,
+                DEFAULT_BATCH_SIZE,
+                Transforms.arrow2python,
+                _reader=reader,
             )
 
         return LOOP.run(do_from_tables())
 
+    def __getstate__(self) -> dict[str, Any]:
+        """Build a picklable state dict for this 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.
+        """
+        permutation_data: Optional[pa.Table] = None
+        if self.permutation_table is not None:
+            permutation_data = self.permutation_table.to_arrow()
+
+        common = {
+            "base_table_name": self.base_table.name,
+            "permutation_data": permutation_data,
+            "split": self.split,
+            "selection": self.selection,
+            "batch_size": self.batch_size,
+            "transform_fn": self.transform_fn,
+            "offset": self.offset,
+            "limit": self.limit,
+            "connection_factory": self.connection_factory,
+        }
+
+        if self.connection_factory is not None:
+            # The factory carries enough state to recover the base table on
+            # its own; we don't need to capture the URI / storage options /
+            # 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,
+        }
+
+    def __setstate__(self, state: dict[str, Any]) -> None:
+        from . import connect
+
+        connection_factory = state["connection_factory"]
+        if connection_factory is not None:
+            base_table = connection_factory(state["base_table_name"])
+        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.
+            mem_db = connect("memory://")
+            base_table = mem_db.create_table(
+                state["base_table_name"], state["base_table_data"]
+            )
+        else:
+            base_db = connect(
+                state["base_table_uri"],
+                storage_options=state["base_table_storage_options"],
+            )
+            base_table = base_db.open_table(
+                state["base_table_name"],
+                namespace_path=state["base_table_namespace"] or None,
+            )
+
+        permutation_table: Optional[LanceTable] = None
+        if state["permutation_data"] is not None:
+            mem_db = connect("memory://")
+            permutation_table = mem_db.create_table(
+                "permutation", state["permutation_data"]
+            )
+
+        self.base_table = base_table
+        self.permutation_table = permutation_table
+        self.split = state["split"]
+        self.selection = state["selection"]
+        self.batch_size = state["batch_size"]
+        self.transform_fn = state["transform_fn"]
+        self.offset = state["offset"]
+        self.limit = state["limit"]
+        self.connection_factory = connection_factory
+        self.reader = LOOP.run(self._build_reader())
+
     @property
     def schema(self) -> pa.Schema:
         async def do_output_schema():
@@ -760,7 +964,9 @@ class Permutation:
         for expensive operations such as image decoding.
         """
         assert transform is not None, "transform is required"
-        return Permutation(self.reader, self.selection, self.batch_size, transform)
+        new = copy.copy(self)
+        new.transform_fn = transform
+        return new
 
     def __getitem__(self, index: int) -> Any:
         """
@@ -795,12 +1001,10 @@ class Permutation:
         """
         Skip the first `skip` rows of the permutation
         """
-
+        new = copy.copy(self)
-        async def do_with_skip():
+        new.offset = skip
-            reader = await self.reader.with_offset(skip)
+        new.reader = LOOP.run(new._build_reader())
-            return self._with_reader(reader)
+        return new
-
-        return LOOP.run(do_with_skip())
 
     @deprecated(details="Use with_take instead")
     def take(self, limit: int) -> "Permutation":
@@ -818,12 +1022,10 @@ class Permutation:
         """
         Limit the permutation to `limit` rows (following any `skip`)
         """
-
+        new = copy.copy(self)
-        async def do_with_take():
+        new.limit = limit
-            reader = await self.reader.with_limit(limit)
+        new.reader = LOOP.run(new._build_reader())
-            return self._with_reader(reader)
+        return new
-
-        return LOOP.run(do_with_take())
 
     @deprecated(details="Use with_repeat instead")
     def repeat(self, times: int) -> "Permutation":

python/python/lancedb/query.py

@@ -25,7 +25,6 @@ import deprecation
 import numpy as np
 import pyarrow as pa
 import pyarrow.compute as pc
-import pyarrow.fs as pa_fs
 import pydantic
 
 from lancedb.pydantic import PYDANTIC_VERSION
@@ -1526,9 +1525,7 @@ class LanceFtsQueryBuilder(LanceQueryBuilder):
         return self._table._output_schema(self.to_query_object())
 
     def to_arrow(self, *, timeout: Optional[timedelta] = None) -> pa.Table:
-        path, fs, exist = self._table._get_fts_index_path()
+        self._table._ensure_no_legacy_fts_index()
-        if exist:
-            return self.tantivy_to_arrow()
 
         query = self._query
         if self._phrase_query:
@@ -1552,90 +1549,6 @@ class LanceFtsQueryBuilder(LanceQueryBuilder):
     ):
         raise NotImplementedError("to_batches on an FTS query")
 
-    def tantivy_to_arrow(self) -> pa.Table:
-        try:
-            import tantivy
-        except ImportError:
-            raise ImportError(
-                "Please install tantivy-py `pip install tantivy` to use the full text search feature."  # noqa: E501
-            )
-
-        from .fts import search_index
-
-        # get the index path
-        path, fs, exist = self._table._get_fts_index_path()
-
-        # check if the index exist
-        if not exist:
-            raise FileNotFoundError(
-                "Fts index does not exist. "
-                "Please first call table.create_fts_index(['<field_names>']) to "
-                "create the fts index."
-            )
-
-        # Check that we are on local filesystem
-        if not isinstance(fs, pa_fs.LocalFileSystem):
-            raise NotImplementedError(
-                "Tantivy-based full text search "
-                "is only supported on the local filesystem"
-            )
-        # open the index
-        index = tantivy.Index.open(path)
-        # get the scores and doc ids
-        query = self._query
-        if self._phrase_query:
-            query = query.replace('"', "'")
-            query = f'"{query}"'
-        limit = self._limit if self._limit is not None else 10
-        row_ids, scores = search_index(
-            index, query, limit, ordering_field=self.ordering_field_name
-        )
-        if len(row_ids) == 0:
-            empty_schema = pa.schema([pa.field("_score", pa.float32())])
-            return pa.Table.from_batches([], schema=empty_schema)
-        scores = pa.array(scores)
-        output_tbl = self._table.to_lance().take(row_ids, columns=self._columns)
-        output_tbl = output_tbl.append_column("_score", scores)
-        # this needs to match vector search results which are uint64
-        row_ids = pa.array(row_ids, type=pa.uint64())
-
-        if self._where is not None:
-            tmp_name = "__lancedb__duckdb__indexer__"
-            output_tbl = output_tbl.append_column(
-                tmp_name, pa.array(range(len(output_tbl)))
-            )
-            try:
-                # TODO would be great to have Substrait generate pyarrow compute
-                # expressions or conversely have pyarrow support SQL expressions
-                # using Substrait
-                import duckdb
-
-                indexer = duckdb.sql(
-                    f"SELECT {tmp_name} FROM output_tbl WHERE {self._where}"
-                ).to_arrow_table()[tmp_name]
-                output_tbl = output_tbl.take(indexer).drop([tmp_name])
-                row_ids = row_ids.take(indexer)
-
-            except ImportError:
-                import tempfile
-
-                import lance
-
-                # TODO Use "memory://" instead once that's supported
-                with tempfile.TemporaryDirectory() as tmp:
-                    ds = lance.write_dataset(output_tbl, tmp)
-                    output_tbl = ds.to_table(filter=self._where)
-                    indexer = output_tbl[tmp_name]
-                    row_ids = row_ids.take(indexer)
-                    output_tbl = output_tbl.drop([tmp_name])
-
-        if self._with_row_id:
-            output_tbl = output_tbl.append_column("_rowid", row_ids)
-
-        if self._reranker is not None:
-            output_tbl = self._reranker.rerank_fts(self._query, output_tbl)
-        return output_tbl
-
     def rerank(self, reranker: Reranker) -> LanceFtsQueryBuilder:
         """Rerank the results using the specified reranker.
 
@@ -1730,7 +1643,7 @@ class LanceHybridQueryBuilder(LanceQueryBuilder):
     def _validate_query(self, query, vector=None, text=None):
         if query is not None and (vector is not None or text is not None):
             raise ValueError(
-                "You can either provide a string query in search() method"
+                "You can either provide a string query in search() method "
                 "or set `vector()` and `text()` explicitly for hybrid search."
                 "But not both."
             )

python/python/lancedb/table.py

@@ -943,29 +943,26 @@ class Table(ABC):
         Parameters
         ----------
         field_names: str or list of str
-            The name(s) of the field to index.
+            The name of the field to index. Native FTS indexes can only be
-            If ``use_tantivy`` is False (default), only a single field name
+            created on a single field at a time. To search over multiple text
-            (str) is supported. To index multiple fields, create a separate
+            fields, create a separate FTS index for each field.
-            FTS index for each field.
         replace: bool, default False
             If True, replace the existing index if it exists. Note that this is
             not yet an atomic operation; the index will be temporarily
             unavailable while the new index is being created.
         writer_heap_size: int, default 1GB
-            Only available with use_tantivy=True
+            Deprecated legacy Tantivy parameter. Any value other than the
+            default raises an error.
         ordering_field_names:
-            A list of unsigned type fields to index to optionally order
+            Deprecated legacy Tantivy parameter. Setting this raises an error.
-            results on at search time.
-            only available with use_tantivy=True
         tokenizer_name: str, default "default"
-            The tokenizer to use for the index. Can be "raw", "default" or the 2 letter
+            A compatibility alias for native tokenizer configs. Can be "raw",
-            language code followed by "_stem". So for english it would be "en_stem".
+            "default" or the 2 letter language code followed by "_stem". So
-            For available languages see: https://docs.rs/tantivy/latest/tantivy/tokenizer/enum.Language.html
+            for english it would be "en_stem".
         use_tantivy: bool, default False
-            If True, use the legacy full-text search implementation based on tantivy.
+            Deprecated legacy Tantivy parameter. Setting this to True raises an
-            If False, use the new full-text search implementation based on lance-index.
+            error.
         with_position: bool, default False
-            Only available with use_tantivy=False
             If False, do not store the positions of the terms in the text.
             This can reduce the size of the index and improve indexing speed.
             But it will raise an exception for phrase queries.
@@ -1746,6 +1743,16 @@ class Table(ABC):
         index_exists = fs.get_file_info(path).type != pa_fs.FileType.NotFound
         return (path, fs, index_exists)
 
+    def _ensure_no_legacy_fts_index(self):
+        path, _, exists = self._get_fts_index_path()
+        if exists:
+            raise ValueError(
+                "Legacy Tantivy FTS index detected at "
+                f"{path}. Tantivy-based FTS has been removed. "
+                "Delete the legacy index and recreate it with "
+                "table.create_fts_index(...)."
+            )
+
     @abstractmethod
     def uses_v2_manifest_paths(self) -> bool:
         """
@@ -2405,84 +2412,63 @@ class LanceTable(Table):
         prefix_only: bool = False,
         name: Optional[str] = None,
     ):
-        if not use_tantivy:
+        self._ensure_no_legacy_fts_index()
-            if not isinstance(field_names, str):
-                raise ValueError(
-                    "Native FTS indexes can only be created on a single field "
-                    "at a time. To search over multiple text fields, create a "
-                    "separate FTS index for each field."
-                )
 
-            if tokenizer_name is None:
+        if use_tantivy:
-                tokenizer_configs = {
+            raise ValueError(
-                    "base_tokenizer": base_tokenizer,
+                "Tantivy-based FTS has been removed. "
-                    "language": language,
+                "Remove use_tantivy and recreate the index with native FTS."
-                    "with_position": with_position,
-                    "max_token_length": max_token_length,
-                    "lower_case": lower_case,
-                    "stem": stem,
-                    "remove_stop_words": remove_stop_words,
-                    "ascii_folding": ascii_folding,
-                    "ngram_min_length": ngram_min_length,
-                    "ngram_max_length": ngram_max_length,
-                    "prefix_only": prefix_only,
-                }
-            else:
-                tokenizer_configs = self.infer_tokenizer_configs(tokenizer_name)
-
-            config = FTS(
-                **tokenizer_configs,
             )
-
+        if ordering_field_names is not None:
-            # delete the existing legacy index if it exists
+            raise ValueError(
-            if replace:
+                "ordering_field_names was only supported by the removed "
-                path, fs, exist = self._get_fts_index_path()
+                "Tantivy-based FTS implementation."
-                if exist:
-                    fs.delete_dir(path)
-
-            LOOP.run(
-                self._table.create_index(
-                    field_names,
-                    replace=replace,
-                    config=config,
-                    name=name,
-                )
             )
-            return
+        if writer_heap_size != 1024 * 1024 * 1024:
-
+            raise ValueError(
-        from .fts import create_index, populate_index
+                "writer_heap_size was only supported by the removed "
-
+                "Tantivy-based FTS implementation."
-        if isinstance(field_names, str):
+            )
-            field_names = [field_names]
+        if not isinstance(field_names, str):
-
+            raise ValueError(
-        if isinstance(ordering_field_names, str):
+                "Native FTS indexes can only be created on a single field "
-            ordering_field_names = [ordering_field_names]
+                "at a time. To search over multiple text fields, create a "
-
+                "separate FTS index for each field."
-        path, fs, exist = self._get_fts_index_path()
+            )
-        if exist:
+        if "." in field_names:
-            if not replace:
+            raise ValueError(
-                raise ValueError("Index already exists. Use replace=True to overwrite.")
+                "Native FTS indexes can only be created on top-level fields. "
-            fs.delete_dir(path)
+                f"Received nested field path: {field_names!r}."
-
-        if not isinstance(fs, pa_fs.LocalFileSystem):
-            raise NotImplementedError(
-                "Full-text search is only supported on the local filesystem"
             )
 
         if tokenizer_name is None:
-            tokenizer_name = "default"
+            tokenizer_configs = {
-        index = create_index(
+                "base_tokenizer": base_tokenizer,
-            path,
+                "language": language,
-            field_names,
+                "with_position": with_position,
-            ordering_fields=ordering_field_names,
+                "max_token_length": max_token_length,
-            tokenizer_name=tokenizer_name,
+                "lower_case": lower_case,
+                "stem": stem,
+                "remove_stop_words": remove_stop_words,
+                "ascii_folding": ascii_folding,
+                "ngram_min_length": ngram_min_length,
+                "ngram_max_length": ngram_max_length,
+                "prefix_only": prefix_only,
+            }
+        else:
+            tokenizer_configs = self.infer_tokenizer_configs(tokenizer_name)
+
+        config = FTS(
+            **tokenizer_configs,
         )
-        populate_index(
+
-            index,
+        LOOP.run(
-            self,
+            self._table.create_index(
-            field_names,
+                field_names,
-            ordering_fields=ordering_field_names,
+                replace=replace,
-            writer_heap_size=writer_heap_size,
+                config=config,
+                name=name,
+            )
         )
 
     @staticmethod

python/python/tests/docs/test_search.py

@@ -180,7 +180,7 @@ def test_fts_fuzzy_query():
         ),
         mode="overwrite",
     )
-    table.create_fts_index("text", use_tantivy=False, replace=True)
+    table.create_fts_index("text", replace=True)
 
     results = table.search(MatchQuery("foo", "text", fuzziness=1)).to_pandas()
     assert len(results) == 4
@@ -230,7 +230,7 @@ def test_fts_boost_query():
         ),
         mode="overwrite",
     )
-    table.create_fts_index("desc", use_tantivy=False, replace=True)
+    table.create_fts_index("desc", replace=True)
 
     results = table.search(
         BoostQuery(
@@ -265,7 +265,7 @@ def test_fts_boolean_query(tmp_path):
         ],
         mode="overwrite",
     )
-    table.create_fts_index("text", use_tantivy=False, replace=True)
+    table.create_fts_index("text", replace=True)
 
     # SHOULD
     results = table.search(
@@ -319,9 +319,7 @@ def test_fts_native():
         ],
     )
 
-    # passing `use_tantivy=False` to use lance FTS index
+    table.create_fts_index("text")
-    # `use_tantivy=True` by default
-    table.create_fts_index("text", use_tantivy=False)
     table.search("puppy").limit(10).select(["text"]).to_list()
     # [{'text': 'Frodo was a happy puppy', '_score': 0.6931471824645996}]
     # ...
@@ -332,7 +330,6 @@ def test_fts_native():
     # --8<-- [start:fts_config_folding]
     table.create_fts_index(
         "text",
-        use_tantivy=False,
         language="French",
         stem=True,
         ascii_folding=True,
@@ -346,7 +343,7 @@ def test_fts_native():
     table.search("puppy").limit(10).where("text='foo'", prefilter=False).to_list()
     # --8<-- [end:fts_postfiltering]
     # --8<-- [start:fts_with_position]
-    table.create_fts_index("text", use_tantivy=False, with_position=True, replace=True)
+    table.create_fts_index("text", with_position=True, replace=True)
     # --8<-- [end:fts_with_position]
     # --8<-- [start:fts_incremental_index]
     table.add([{"vector": [3.1, 4.1], "text": "Frodo was a happy puppy"}])

python/python/tests/test_db.py

@@ -15,8 +15,7 @@ import pytest
 from lancedb.pydantic import LanceModel, Vector
 
 
-@pytest.mark.parametrize("use_tantivy", [True, False])
+def test_basic(tmp_path):
-def test_basic(tmp_path, use_tantivy):
     db = lancedb.connect(tmp_path)
 
     assert db.uri == str(tmp_path)
@@ -49,7 +48,7 @@ def test_basic(tmp_path, use_tantivy):
     assert len(rs) == 1
     assert rs["item"].iloc[0] == "foo"
 
-    table.create_fts_index("item", use_tantivy=use_tantivy)
+    table.create_fts_index("item")
     rs = table.search("bar", query_type="fts").to_pandas()
     assert len(rs) == 1
     assert rs["item"].iloc[0] == "bar"

python/python/tests/test_fts.py

@@ -36,9 +36,6 @@ import pytest
 import pytest_asyncio
 from utils import exception_output
 
-pytest.importorskip("lancedb.fts")
-tantivy = pytest.importorskip("tantivy")
-
 
 @pytest.fixture
 def table(tmp_path) -> ldb.table.LanceTable:
@@ -144,58 +141,53 @@ async def async_table(tmp_path) -> ldb.table.AsyncTable:
     return table
 
 
-def test_create_index(tmp_path):
+@pytest.mark.parametrize(
-    index = ldb.fts.create_index(str(tmp_path / "index"), ["text"])
+    ("kwargs", "match"),
-    assert isinstance(index, tantivy.Index)
+    [
-    assert os.path.exists(str(tmp_path / "index"))
+        (
+            {"use_tantivy": True},
+            "Tantivy-based FTS has been removed",
+        ),
+        (
+            {"ordering_field_names": ["count"]},
+            "ordering_field_names was only supported",
+        ),
+        (
+            {"writer_heap_size": 128},
+            "writer_heap_size was only supported",
+        ),
+    ],
+)
+def test_reject_removed_tantivy_parameters(table, kwargs, match):
+    with pytest.raises(ValueError, match=match):
+        table.create_fts_index("text", **kwargs)
 
 
-def test_create_index_with_stemming(tmp_path, table):
+def test_reject_legacy_tantivy_index(table):
-    index = ldb.fts.create_index(
+    path, _, _ = table._get_fts_index_path()
-        str(tmp_path / "index"), ["text"], tokenizer_name="en_stem"
+    os.makedirs(path, exist_ok=True)
-    )
-    assert isinstance(index, tantivy.Index)
-    assert os.path.exists(str(tmp_path / "index"))
 
-    # Check stemming by running tokenizer on non empty table
+    with pytest.raises(ValueError, match="Legacy Tantivy FTS index detected"):
-    table.create_fts_index("text", tokenizer_name="en_stem", use_tantivy=True)
+        table.search("puppy").limit(5).to_list()
+
+    with pytest.raises(ValueError, match="Legacy Tantivy FTS index detected"):
+        table.create_fts_index("text")
 
 
-@pytest.mark.parametrize("use_tantivy", [True, False])
 @pytest.mark.parametrize("with_position", [True, False])
-def test_create_inverted_index(table, use_tantivy, with_position):
+def test_create_inverted_index(table, with_position):
-    if use_tantivy and not with_position:
-        pytest.skip("we don't support building a tantivy index without position")
     table.create_fts_index(
         "text",
-        use_tantivy=use_tantivy,
         with_position=with_position,
         name="custom_fts_index",
     )
-    if not use_tantivy:
+    indices = table.list_indices()
-        indices = table.list_indices()
+    fts_indices = [i for i in indices if i.index_type == "FTS"]
-        fts_indices = [i for i in indices if i.index_type == "FTS"]
+    assert any(i.name == "custom_fts_index" for i in fts_indices)
-        assert any(i.name == "custom_fts_index" for i in fts_indices)
 
 
-def test_populate_index(tmp_path, table):
+def test_search_fts(table):
-    index = ldb.fts.create_index(str(tmp_path / "index"), ["text"])
+    table.create_fts_index("text")
-    assert ldb.fts.populate_index(index, table, ["text"]) == len(table)
-
-
-def test_search_index(tmp_path, table):
-    index = ldb.fts.create_index(str(tmp_path / "index"), ["text"])
-    ldb.fts.populate_index(index, table, ["text"])
-    index.reload()
-    results = ldb.fts.search_index(index, query="puppy", limit=5)
-    assert len(results) == 2
-    assert len(results[0]) == 5  # row_ids
-    assert len(results[1]) == 5  # _score
-
-
-@pytest.mark.parametrize("use_tantivy", [True, False])
-def test_search_fts(table, use_tantivy):
-    table.create_fts_index("text", use_tantivy=use_tantivy)
     results = table.search("puppy").select(["id", "text"]).limit(5).to_list()
     assert len(results) == 5
     assert len(results[0]) == 3  # id, text, _score
@@ -204,53 +196,52 @@ def test_search_fts(table, use_tantivy):
     results = table.search("puppy").select(["id", "text"]).to_list()
     assert len(results) == 10
 
-    if not use_tantivy:
+    # Test with a query
-        # Test with a query
+    results = (
-        results = (
+        table.search(MatchQuery("puppy", "text"))
-            table.search(MatchQuery("puppy", "text"))
+        .select(["id", "text"])
-            .select(["id", "text"])
+        .limit(5)
-            .limit(5)
+        .to_list()
-            .to_list()
+    )
-        )
+    assert len(results) == 5
-        assert len(results) == 5
 
-        # Test boost query
+    # Test boost query
-        results = (
+    results = (
-            table.search(
+        table.search(
-                BoostQuery(
+            BoostQuery(
-                    MatchQuery("puppy", "text"),
+                MatchQuery("puppy", "text"),
-                    MatchQuery("runs", "text"),
+                MatchQuery("runs", "text"),
-                )
             )
-            .select(["id", "text"])
-            .limit(5)
-            .to_list()
         )
-        assert len(results) == 5
+        .select(["id", "text"])
+        .limit(5)
+        .to_list()
+    )
+    assert len(results) == 5
 
-        # Test multi match query
+    # Test multi match query
-        table.create_fts_index("text2", use_tantivy=use_tantivy)
+    table.create_fts_index("text2")
-        results = (
+    results = (
-            table.search(MultiMatchQuery("puppy", ["text", "text2"]))
+        table.search(MultiMatchQuery("puppy", ["text", "text2"]))
-            .select(["id", "text"])
+        .select(["id", "text"])
-            .limit(5)
+        .limit(5)
-            .to_list()
+        .to_list()
-        )
+    )
-        assert len(results) == 5
+    assert len(results) == 5
-        assert len(results[0]) == 3  # id, text, _score
+    assert len(results[0]) == 3  # id, text, _score
 
-        # Test boolean query
+    # Test boolean query
-        results = (
+    results = (
-            table.search(MatchQuery("puppy", "text") & MatchQuery("runs", "text"))
+        table.search(MatchQuery("puppy", "text") & MatchQuery("runs", "text"))
-            .select(["id", "text"])
+        .select(["id", "text"])
-            .limit(5)
+        .limit(5)
-            .to_list()
+        .to_list()
-        )
+    )
-        assert len(results) == 5
+    assert len(results) == 5
-        assert len(results[0]) == 3  # id, text, _score
+    assert len(results[0]) == 3  # id, text, _score
-        for r in results:
+    for r in results:
-            assert "puppy" in r["text"]
+        assert "puppy" in r["text"]
-            assert "runs" in r["text"]
+        assert "runs" in r["text"]
 
 
 @pytest.mark.asyncio
@@ -318,13 +309,13 @@ async def test_fts_select_async(async_table):
 
 
 def test_search_fts_phrase_query(table):
-    table.create_fts_index("text", use_tantivy=False, with_position=False)
+    table.create_fts_index("text", with_position=False)
     try:
         phrase_results = table.search('"puppy runs"').limit(100).to_list()
         assert False
     except Exception:
         pass
-    table.create_fts_index("text", use_tantivy=False, with_position=True, replace=True)
+    table.create_fts_index("text", with_position=True, replace=True)
     results = table.search("puppy").limit(100).to_list()
 
     # Test with quotation marks
@@ -375,8 +366,8 @@ async def test_search_fts_phrase_query_async(async_table):
 
 
 def test_search_fts_specify_column(table):
-    table.create_fts_index("text", use_tantivy=False)
+    table.create_fts_index("text")
-    table.create_fts_index("text2", use_tantivy=False)
+    table.create_fts_index("text2")
 
     results = table.search("puppy", fts_columns="text").limit(5).to_list()
     assert len(results) == 5
@@ -470,42 +461,8 @@ async def test_search_fts_specify_column_async(async_table):
         pass
 
 
-def test_search_ordering_field_index_table(tmp_path, table):
+def test_create_index_from_table(tmp_path, table):
-    table.create_fts_index("text", ordering_field_names=["count"], use_tantivy=True)
+    table.create_fts_index("text")
-    rows = (
-        table.search("puppy", ordering_field_name="count")
-        .limit(20)
-        .select(["text", "count"])
-        .to_list()
-    )
-    for r in rows:
-        assert "puppy" in r["text"]
-    assert sorted(rows, key=lambda x: x["count"], reverse=True) == rows
-
-
-def test_search_ordering_field_index(tmp_path, table):
-    index = ldb.fts.create_index(
-        str(tmp_path / "index"), ["text"], ordering_fields=["count"]
-    )
-
-    ldb.fts.populate_index(index, table, ["text"], ordering_fields=["count"])
-    index.reload()
-    results = ldb.fts.search_index(
-        index, query="puppy", limit=5, ordering_field="count"
-    )
-    assert len(results) == 2
-    assert len(results[0]) == 5  # row_ids
-    assert len(results[1]) == 5  # _distance
-    rows = table.to_lance().take(results[0]).to_pylist()
-
-    for r in rows:
-        assert "puppy" in r["text"]
-    assert sorted(rows, key=lambda x: x["count"], reverse=True) == rows
-
-
-@pytest.mark.parametrize("use_tantivy", [True, False])
-def test_create_index_from_table(tmp_path, table, use_tantivy):
-    table.create_fts_index("text", use_tantivy=use_tantivy)
     df = table.search("puppy").limit(5).select(["text"]).to_pandas()
     assert len(df) <= 5
     assert "text" in df.columns
@@ -525,36 +482,24 @@ def test_create_index_from_table(tmp_path, table, use_tantivy):
     )
 
     with pytest.raises(Exception, match="already exists"):
-        table.create_fts_index("text", use_tantivy=use_tantivy)
+        table.create_fts_index("text")
 
-    table.create_fts_index("text", replace=True, use_tantivy=use_tantivy)
+    table.create_fts_index("text", replace=True)
     assert len(table.search("gorilla").limit(1).to_pandas()) == 1
 
 
 def test_create_index_multiple_columns(tmp_path, table):
-    table.create_fts_index(["text", "text2"], use_tantivy=True)
+    with pytest.raises(ValueError, match="Native FTS indexes can only be created"):
-    df = table.search("puppy").limit(5).to_pandas()
+        table.create_fts_index(["text", "text2"])
-    assert len(df) == 5
-    assert "text" in df.columns
-    assert "text2" in df.columns
-
-
-def test_empty_rs(tmp_path, table, mocker):
-    table.create_fts_index(["text", "text2"], use_tantivy=True)
-    mocker.patch("lancedb.fts.search_index", return_value=([], []))
-    df = table.search("puppy").limit(5).to_pandas()
-    assert len(df) == 0
 
 
 def test_nested_schema(tmp_path, table):
-    table.create_fts_index("nested.text", use_tantivy=True)
+    with pytest.raises(ValueError, match="top-level fields"):
-    rs = table.search("puppy").limit(5).to_list()
+        table.create_fts_index("nested.text")
-    assert len(rs) == 5
 
 
-@pytest.mark.parametrize("use_tantivy", [True, False])
+def test_search_index_with_filter(table):
-def test_search_index_with_filter(table, use_tantivy):
+    table.create_fts_index("text")
-    table.create_fts_index("text", use_tantivy=use_tantivy)
     orig_import = __import__
 
     def import_mock(name, *args):
@@ -584,8 +529,7 @@ def test_search_index_with_filter(table, use_tantivy):
         assert r["_rowid"] is not None
 
 
-@pytest.mark.parametrize("use_tantivy", [True, False])
+def test_null_input(table):
-def test_null_input(table, use_tantivy):
     table.add(
         [
             {
@@ -598,14 +542,13 @@ def test_null_input(table, use_tantivy):
             }
         ]
     )
-    table.create_fts_index("text", use_tantivy=use_tantivy)
+    table.create_fts_index("text")
 
 
 def test_syntax(table):
     # https://github.com/lancedb/lancedb/issues/769
-    table.create_fts_index("text", use_tantivy=True)
+    table.create_fts_index("text")
-    with pytest.raises(ValueError, match="Syntax Error"):
+    table.search("they could have been dogs OR").limit(10).to_list()
-        table.search("they could have been dogs OR").limit(10).to_list()
 
     # these should work
 
@@ -616,6 +559,7 @@ def test_syntax(table):
     ).to_list()
 
     # phrase queries
+    table.create_fts_index("text", with_position=True, replace=True)
     table.search("they could have been dogs OR cats").phrase_query().limit(10).to_list()
     table.search('"they could have been dogs OR cats"').limit(10).to_list()
     table.search('''"the cats OR dogs were not really 'pets' at all"''').limit(
@@ -639,7 +583,7 @@ def test_language(mem_db: DBConnection):
     table = mem_db.create_table("test", data=data)
 
     with pytest.raises(ValueError) as e:
-        table.create_fts_index("text", use_tantivy=False, language="klingon")
+        table.create_fts_index("text", language="klingon")
 
     assert exception_output(e) == (
         "ValueError: LanceDB does not support the requested language: 'klingon'\n"
@@ -650,7 +594,6 @@ def test_language(mem_db: DBConnection):
 
     table.create_fts_index(
         "text",
-        use_tantivy=False,
         language="French",
         stem=True,
         ascii_folding=True,
@@ -690,7 +633,7 @@ def test_fts_on_list(mem_db: DBConnection):
         }
     )
     table = mem_db.create_table("test", data=data)
-    table.create_fts_index("text", use_tantivy=False, with_position=True)
+    table.create_fts_index("text", with_position=True)
 
     res = table.search("lance").limit(5).to_list()
     assert len(res) == 3
@@ -702,7 +645,7 @@ def test_fts_on_list(mem_db: DBConnection):
 def test_fts_ngram(mem_db: DBConnection):
     data = pa.table({"text": ["hello world", "lance database", "lance is cool"]})
     table = mem_db.create_table("test", data=data)
-    table.create_fts_index("text", use_tantivy=False, base_tokenizer="ngram")
+    table.create_fts_index("text", base_tokenizer="ngram")
 
     results = table.search("lan", query_type="fts").limit(10).to_list()
     assert len(results) == 2
@@ -721,7 +664,6 @@ def test_fts_ngram(mem_db: DBConnection):
     # test setting min_ngram_length and prefix_only
     table.create_fts_index(
         "text",
-        use_tantivy=False,
         base_tokenizer="ngram",
         replace=True,
         ngram_min_length=2,
@@ -886,7 +828,7 @@ def test_fts_query_to_json():
 
 
 def test_fts_fast_search(table):
-    table.create_fts_index("text", use_tantivy=False)
+    table.create_fts_index("text")
 
     # Insert some unindexed data
     table.add(

python/python/tests/test_hybrid_query.py

@@ -28,7 +28,7 @@ def sync_table(tmpdir_factory) -> Table:
         }
     )
     table = db.create_table("test", data)
-    table.create_fts_index("text", with_position=False, use_tantivy=False)
+    table.create_fts_index("text", with_position=False)
     return table
 
 
@@ -192,7 +192,7 @@ def table_with_id(tmpdir_factory) -> Table:
         }
     )
     table = db.create_table("test_with_id", data)
-    table.create_fts_index("text", with_position=False, use_tantivy=False)
+    table.create_fts_index("text", with_position=False)
     return table
 
 

python/python/tests/test_permutation.py

@@ -9,21 +9,6 @@ from lancedb import DBConnection, Table, connect
 from lancedb.permutation import Permutation, Permutations, permutation_builder
 
 
-def test_permutation_persistence(tmp_path):
-    db = connect(tmp_path)
-    tbl = db.create_table("test_table", pa.table({"x": range(100), "y": range(100)}))
-
-    permutation_tbl = (
-        permutation_builder(tbl).shuffle().persist(db, "test_permutation").execute()
-    )
-    assert permutation_tbl.count_rows() == 100
-
-    re_open = db.open_table("test_permutation")
-    assert re_open.count_rows() == 100
-
-    assert permutation_tbl.to_arrow() == re_open.to_arrow()
-
-
 def test_split_random_ratios(mem_db):
     """Test random splitting with ratios."""
     tbl = mem_db.create_table(

python/python/tests/test_query.py

@@ -1385,7 +1385,7 @@ def test_query_timeout(tmp_path):
         }
     )
     table = db.create_table("test", data)
-    table.create_fts_index("text", use_tantivy=False)
+    table.create_fts_index("text")
 
     with pytest.raises(Exception, match="Query timeout"):
         table.search().where("text = 'a'").to_list(timeout=timedelta(0))

python/python/tests/test_remote_db.py

@@ -6,6 +6,8 @@ import contextlib
 from datetime import timedelta
 import http.server
 import json
+import multiprocessing as mp
+import sys
 import threading
 import time
 from unittest.mock import MagicMock, patch
@@ -1230,3 +1232,82 @@ def test_background_loop_cancellation(exception):
         with pytest.raises(exception):
             loop.run(None)
         mock_future.cancel.assert_called_once()
+
+
+def _remote_fork_child(port: int, queue) -> None:
+    # Build a fresh Connection in the child so we exercise the at-fork-child
+    # tokio runtime reset rather than relying on an inherited reqwest client.
+    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},
+        },
+    )
+    queue.put(db.table_names())
+
+
+@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_connection_after_fork():
+    """A freshly-built remote Connection in a forked child should not hang.
+
+    The pyo3-async-runtimes tokio runtime would otherwise be inherited from
+    the parent with dead worker threads; the at-fork-child handler in our
+    runtime module rebuilds it on first use in the child.
+    """
+
+    def handler(request):
+        request.send_response(200)
+        request.send_header("Content-Type", "application/json")
+        request.end_headers()
+        request.wfile.write(b'{"tables": []}')
+
+    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:
+        # Hit the server in the parent first so the runtime + LOOP are warm
+        # before fork; a fresh child must still succeed.
+        parent_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},
+            },
+        )
+        assert parent_db.table_names() == []
+
+        ctx = mp.get_context("fork")
+        queue = ctx.Queue()
+        proc = ctx.Process(target=_remote_fork_child, args=(port, 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 connection 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() == []
+
+        # Parent connection must still be usable after the child returned.
+        assert parent_db.table_names() == []
+    finally:
+        server.shutdown()
+        server_thread.join()

python/python/tests/test_rerankers.py

@@ -26,11 +26,8 @@ from lancedb.rerankers import (
 )
 from lancedb.table import LanceTable
 
-# Tests rely on FTS index
-pytest.importorskip("lancedb.fts")
 
-
+def get_test_table(tmp_path):
-def get_test_table(tmp_path, use_tantivy):
     db = lancedb.connect(tmp_path)
     # Create a LanceDB table schema with a vector and a text column
     emb = EmbeddingFunctionRegistry.get_instance().get("test").create()
@@ -98,7 +95,7 @@ def get_test_table(tmp_path, use_tantivy):
     )
 
     # Create a fts index
-    table.create_fts_index("text", use_tantivy=use_tantivy, replace=True)
+    table.create_fts_index("text", replace=True)
 
     return table, MyTable
 
@@ -208,8 +205,8 @@ def _run_test_reranker(reranker, table, query, query_vector, schema):
     assert len(result) == 20 and result == result_arrow
 
 
-def _run_test_hybrid_reranker(reranker, tmp_path, use_tantivy):
+def _run_test_hybrid_reranker(reranker, tmp_path):
-    table, schema = get_test_table(tmp_path, use_tantivy)
+    table, schema = get_test_table(tmp_path)
     # The default reranker
     result1 = (
         table.search(
@@ -285,8 +282,7 @@ def _run_test_hybrid_reranker(reranker, tmp_path, use_tantivy):
     )
 
 
-@pytest.mark.parametrize("use_tantivy", [True, False])
+def test_linear_combination(tmp_path):
-def test_linear_combination(tmp_path, use_tantivy):
     reranker = LinearCombinationReranker()
 
     vector_results = pa.Table.from_pydict(
@@ -313,22 +309,20 @@ def test_linear_combination(tmp_path, use_tantivy):
     assert "_score" not in combined_results.column_names
     assert "_relevance_score" in combined_results.column_names
 
-    _run_test_hybrid_reranker(reranker, tmp_path, use_tantivy)
+    _run_test_hybrid_reranker(reranker, tmp_path)
 
 
-@pytest.mark.parametrize("use_tantivy", [True, False])
+def test_rrf_reranker(tmp_path):
-def test_rrf_reranker(tmp_path, use_tantivy):
     reranker = RRFReranker()
-    _run_test_hybrid_reranker(reranker, tmp_path, use_tantivy)
+    _run_test_hybrid_reranker(reranker, tmp_path)
 
 
-@pytest.mark.parametrize("use_tantivy", [True, False])
+def test_mrr_reranker(tmp_path):
-def test_mrr_reranker(tmp_path, use_tantivy):
     reranker = MRRReranker()
-    _run_test_hybrid_reranker(reranker, tmp_path, use_tantivy)
+    _run_test_hybrid_reranker(reranker, tmp_path)
 
     # Test multi-vector part
-    table, schema = get_test_table(tmp_path, use_tantivy)
+    table, schema = get_test_table(tmp_path)
     query = "single player experience"
     rs1 = table.search(query, vector_column_name="vector").limit(10).with_row_id(True)
     rs2 = (
@@ -363,7 +357,7 @@ def test_rrf_reranker_distance():
     table = db.create_table("test", data)
 
     table.create_index(num_partitions=1, num_sub_vectors=2)
-    table.create_fts_index("text", use_tantivy=False)
+    table.create_fts_index("text")
 
     reranker = RRFReranker(return_score="all")
 
@@ -422,35 +416,31 @@ def test_rrf_reranker_distance():
 @pytest.mark.skipif(
     os.environ.get("COHERE_API_KEY") is None, reason="COHERE_API_KEY not set"
 )
-@pytest.mark.parametrize("use_tantivy", [True, False])
+def test_cohere_reranker(tmp_path):
-def test_cohere_reranker(tmp_path, use_tantivy):
     pytest.importorskip("cohere")
     reranker = CohereReranker()
-    table, schema = get_test_table(tmp_path, use_tantivy)
+    table, schema = get_test_table(tmp_path)
     _run_test_reranker(reranker, table, "single player experience", None, schema)
 
 
-@pytest.mark.parametrize("use_tantivy", [True, False])
+def test_cross_encoder_reranker(tmp_path):
-def test_cross_encoder_reranker(tmp_path, use_tantivy):
     pytest.importorskip("sentence_transformers")
     reranker = CrossEncoderReranker()
-    table, schema = get_test_table(tmp_path, use_tantivy)
+    table, schema = get_test_table(tmp_path)
     _run_test_reranker(reranker, table, "single player experience", None, schema)
 
 
-@pytest.mark.parametrize("use_tantivy", [True, False])
+def test_colbert_reranker(tmp_path):
-def test_colbert_reranker(tmp_path, use_tantivy):
     pytest.importorskip("rerankers")
     reranker = ColbertReranker()
-    table, schema = get_test_table(tmp_path, use_tantivy)
+    table, schema = get_test_table(tmp_path)
     _run_test_reranker(reranker, table, "single player experience", None, schema)
 
 
-@pytest.mark.parametrize("use_tantivy", [True, False])
+def test_answerdotai_reranker(tmp_path):
-def test_answerdotai_reranker(tmp_path, use_tantivy):
     pytest.importorskip("rerankers")
     reranker = AnswerdotaiRerankers()
-    table, schema = get_test_table(tmp_path, use_tantivy)
+    table, schema = get_test_table(tmp_path)
     _run_test_reranker(reranker, table, "single player experience", None, schema)
 
 
@@ -459,10 +449,9 @@ def test_answerdotai_reranker(tmp_path, use_tantivy):
     or os.environ.get("OPENAI_BASE_URL") is not None,
     reason="OPENAI_API_KEY not set",
 )
-@pytest.mark.parametrize("use_tantivy", [True, False])
+def test_openai_reranker(tmp_path):
-def test_openai_reranker(tmp_path, use_tantivy):
     pytest.importorskip("openai")
-    table, schema = get_test_table(tmp_path, use_tantivy)
+    table, schema = get_test_table(tmp_path)
     reranker = OpenaiReranker()
     _run_test_reranker(reranker, table, "single player experience", None, schema)
 
@@ -470,10 +459,9 @@ def test_openai_reranker(tmp_path, use_tantivy):
 @pytest.mark.skipif(
     os.environ.get("JINA_API_KEY") is None, reason="JINA_API_KEY not set"
 )
-@pytest.mark.parametrize("use_tantivy", [True, False])
+def test_jina_reranker(tmp_path):
-def test_jina_reranker(tmp_path, use_tantivy):
     pytest.importorskip("jina")
-    table, schema = get_test_table(tmp_path, use_tantivy)
+    table, schema = get_test_table(tmp_path)
     reranker = JinaReranker()
     _run_test_reranker(reranker, table, "single player experience", None, schema)
 
@@ -481,11 +469,10 @@ def test_jina_reranker(tmp_path, use_tantivy):
 @pytest.mark.skipif(
     os.environ.get("VOYAGE_API_KEY") is None, reason="VOYAGE_API_KEY not set"
 )
-@pytest.mark.parametrize("use_tantivy", [True, False])
+def test_voyageai_reranker(tmp_path):
-def test_voyageai_reranker(tmp_path, use_tantivy):
     pytest.importorskip("voyageai")
     reranker = VoyageAIReranker(model_name="rerank-2.5")
-    table, schema = get_test_table(tmp_path, use_tantivy)
+    table, schema = get_test_table(tmp_path)
     _run_test_reranker(reranker, table, "single player experience", None, schema)
 
 
@@ -504,7 +491,7 @@ def test_empty_result_reranker():
 
     # Create empty table with schema
     empty_table = db.create_table("empty_table", schema=schema, mode="overwrite")
-    empty_table.create_fts_index("text", use_tantivy=False, replace=True)
+    empty_table.create_fts_index("text", replace=True)
     for reranker in [
         CrossEncoderReranker(),
         # ColbertReranker(),
@@ -603,11 +590,10 @@ def test_empty_hybrid_result_reranker():
     assert "_rowid" in result.column_names
 
 
-@pytest.mark.parametrize("use_tantivy", [True, False])
+def test_cross_encoder_reranker_return_all(tmp_path):
-def test_cross_encoder_reranker_return_all(tmp_path, use_tantivy):
     pytest.importorskip("sentence_transformers")
     reranker = CrossEncoderReranker(return_score="all")
-    table, schema = get_test_table(tmp_path, use_tantivy)
+    table, schema = get_test_table(tmp_path)
     query = "single player experience"
     result = (
         table.search(query, query_type="hybrid", vector_column_name="vector")

python/python/tests/test_s3.py

@@ -242,8 +242,8 @@ def test_s3_dynamodb_sync(s3_bucket: str, commit_table: str, monkeypatch):
 
     # FTS indices should error since they are not supported yet.
     with pytest.raises(
-        NotImplementedError,
+        ValueError,
-        match="Full-text search is only supported on the local filesystem",
+        match="Tantivy-based FTS has been removed",
     ):
         table.create_fts_index("x", use_tantivy=True)
 

python/python/tests/test_table.py

@@ -1948,7 +1948,6 @@ def setup_hybrid_search_table(db: DBConnection, embedding_func):
 
 def test_hybrid_search(tmp_db: DBConnection):
     # This test uses an FTS index
-    pytest.importorskip("lancedb.fts")
     pytest.importorskip("lance")
 
     table, MyTable, emb = setup_hybrid_search_table(tmp_db, "test")
@@ -2019,7 +2018,6 @@ def test_hybrid_search(tmp_db: DBConnection):
 
 def test_hybrid_search_metric_type(tmp_db: DBConnection):
     # This test uses an FTS index
-    pytest.importorskip("lancedb.fts")
     pytest.importorskip("lance")
 
     # Need to use nonnorm as the embedding function so l2 and dot results

python/python/tests/test_torch.py

@@ -1,14 +1,29 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright The LanceDB Authors
 
+import functools
+import multiprocessing as mp
+import pickle
+import sys
+
+import lancedb
 import pyarrow as pa
 import pytest
+from lancedb.permutation import Permutation, Permutations, permutation_builder
 from lancedb.util import tbl_to_tensor
-from lancedb.permutation import Permutation
 
 torch = pytest.importorskip("torch")
 
 
+def _open_native_table(uri: str, table_name: str):
+    """Top-level connection factory used by the explicit-factory pickle test.
+
+    Defined at module scope so that pickle can resolve it by name in the
+    worker / unpickling process.
+    """
+    return lancedb.connect(uri).open_table(table_name)
+
+
 def test_table_dataloader(mem_db):
     table = mem_db.create_table("test_table", pa.table({"a": range(1000)}))
     dataloader = torch.utils.data.DataLoader(
@@ -40,3 +55,156 @@ def test_permutation_dataloader(mem_db):
     for batch in dataloader:
         assert batch.size(0) == 1
         assert batch.size(1) == 10
+
+
+def test_permutation_is_picklable(tmp_db):
+    """A Permutation must be picklable so it can be used with PyTorch's
+    DataLoader when num_workers > 0 (which uses multiprocessing and pickles
+    the dataset to pass it to worker processes)."""
+    table = tmp_db.create_table("test_table", pa.table({"a": range(1000)}))
+    permutation = Permutation.identity(table)
+
+    pickled = pickle.dumps(permutation)
+    restored = pickle.loads(pickled)
+
+    assert len(restored) == 1000
+    rows = restored.__getitems__([0, 1, 2])
+    assert rows == [{"a": 0}, {"a": 1}, {"a": 2}]
+
+
+def test_permutation_with_memory_base_is_picklable(mem_db):
+    """An in-memory base table is inlined into the pickle as Arrow IPC bytes
+    and rebuilt on the other side as an in-memory LanceTable, so the
+    Permutation round-trips even though the original database can't be
+    reopened across processes."""
+    table = mem_db.create_table("test_table", pa.table({"a": range(50)}))
+    permutation = Permutation.identity(table)
+
+    restored = pickle.loads(pickle.dumps(permutation))
+
+    assert len(restored) == 50
+    assert restored.__getitems__([0, 10, 49]) == [{"a": 0}, {"a": 10}, {"a": 49}]
+
+
+def test_permutation_dataloader_multiprocessing(tmp_db):
+    """Using a Permutation with a PyTorch DataLoader that has num_workers > 0
+    must work end-to-end. Each worker process gets a pickled copy of the
+    dataset and reads batches from it."""
+    table = tmp_db.create_table("test_table", pa.table({"a": range(1000)}))
+    permutation = Permutation.identity(table)
+
+    dataloader = torch.utils.data.DataLoader(
+        permutation,
+        batch_size=10,
+        shuffle=True,
+        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 == 1000
+
+
+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
+    for remote / cloud connections where the URI alone isn't reopenable."""
+    db = lancedb.connect(tmp_path)
+    db.create_table("test_table", pa.table({"a": range(50)}))
+
+    factory = functools.partial(_open_native_table, str(tmp_path))
+    permutation = Permutation.identity(factory("test_table")).with_connection_factory(
+        factory
+    )
+
+    restored = pickle.loads(pickle.dumps(permutation))
+
+    assert len(restored) == 50
+    # The factory survives pickling and is what powered base-table reopen.
+    assert restored.connection_factory is not None
+    assert restored.connection_factory.func is _open_native_table
+    assert restored.__getitems__([0, 1, 2]) == [{"a": 0}, {"a": 1}, {"a": 2}]
+
+
+def test_permutation_with_builder_is_picklable(tmp_db):
+    """A Permutation built from a non-identity permutation table must round-trip
+    through pickle while preserving the row order defined by the permutation."""
+    table = tmp_db.create_table("test_table", pa.table({"a": range(100)}))
+    perm_tbl = (
+        permutation_builder(table)
+        .split_random(ratios=[0.8, 0.2], seed=42, split_names=["train", "test"])
+        .shuffle(seed=42)
+        .execute()
+    )
+    permutations = Permutations(table, perm_tbl)
+    permutation = permutations["train"]
+
+    indices = list(range(len(permutation)))
+    expected = permutation.__getitems__(indices)
+
+    restored = pickle.loads(pickle.dumps(permutation))
+
+    assert len(restored) == len(permutation)
+    assert restored.__getitems__(indices) == expected
+
+
+def _multiworker_dataloader_target(db_uri: str, result_queue):
+    import lancedb
+    from lancedb.permutation import Permutation
+
+    db = lancedb.connect(db_uri)
+    table = db.open_table("test_table")
+    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=(
+        "fork() is unavailable on Windows and unsafe on macOS "
+        "(Apple frameworks/TLS are not fork-safe)"
+    ),
+)
+def test_permutation_dataloader_fork_workers(tmp_path):
+    """A Permutation used by a fork-based DataLoader should not hang.
+
+    PyTorch's DataLoader uses fork-based multiprocessing by default on Linux.
+    LanceDB drives async work through a background asyncio thread that does
+    not survive a fork, so any LOOP.run() in a worker blocks forever.
+    """
+    import lancedb
+
+    db_uri = str(tmp_path / "db")
+    db = lancedb.connect(db_uri)
+    db.create_table("test_table", pa.table({"a": list(range(1000))}))
+
+    ctx = mp.get_context("spawn")
+    queue = ctx.Queue()
+    proc = ctx.Process(target=_multiworker_dataloader_target, args=(db_uri, 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("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() == 100

python/src/arrow.rs

@@ -3,6 +3,8 @@
 
 use std::sync::Arc;
 
+use crate::error::PythonErrorExt;
+use crate::runtime::future_into_py;
 use arrow::{
     datatypes::SchemaRef,
     pyarrow::{IntoPyArrow, ToPyArrow},
@@ -12,9 +14,6 @@ use lancedb::arrow::SendableRecordBatchStream;
 use pyo3::{
     Bound, Py, PyAny, PyRef, PyResult, Python, exceptions::PyStopAsyncIteration, pyclass, pymethods,
 };
-use pyo3_async_runtimes::tokio::future_into_py;
-
-use crate::error::PythonErrorExt;
 
 #[pyclass]
 pub struct RecordBatchStream {

python/src/connection.rs

@@ -7,6 +7,12 @@ use std::{
     time::Duration,
 };
 
+use crate::{
+    error::PythonErrorExt,
+    namespace::{create_namespace_storage_options_provider, extract_namespace_arc},
+    runtime::future_into_py,
+    table::Table,
+};
 use arrow::{datatypes::Schema, ffi_stream::ArrowArrayStreamReader, pyarrow::FromPyArrow};
 use lancedb::{
     connection::Connection as LanceConnection,
@@ -20,13 +26,6 @@ use pyo3::{
     pyclass, pyfunction, pymethods,
     types::{PyDict, PyDictMethods},
 };
-use pyo3_async_runtimes::tokio::future_into_py;
-
-use crate::{
-    error::PythonErrorExt,
-    namespace::{create_namespace_storage_options_provider, extract_namespace_arc},
-    table::Table,
-};
 
 #[pyclass]
 pub struct Connection {
@@ -525,7 +524,7 @@ impl Connection {
 }
 
 #[pyfunction]
-#[pyo3(signature = (uri, api_key=None, region=None, host_override=None, read_consistency_interval=None, client_config=None, storage_options=None, session=None))]
+#[pyo3(signature = (uri, api_key=None, region=None, host_override=None, read_consistency_interval=None, client_config=None, storage_options=None, session=None, manifest_enabled=false, namespace_client_properties=None))]
 #[allow(clippy::too_many_arguments)]
 pub fn connect(
     py: Python<'_>,
@@ -537,6 +536,8 @@ pub fn connect(
     client_config: Option<PyClientConfig>,
     storage_options: Option<HashMap<String, String>>,
     session: Option<crate::session::Session>,
+    manifest_enabled: bool,
+    namespace_client_properties: Option<HashMap<String, String>>,
 ) -> PyResult<Bound<'_, PyAny>> {
     future_into_py(py, async move {
         let mut builder = lancedb::connect(&uri);
@@ -556,6 +557,12 @@ pub fn connect(
         if let Some(storage_options) = storage_options {
             builder = builder.storage_options(storage_options);
         }
+        if manifest_enabled {
+            builder = builder.manifest_enabled(true);
+        }
+        if let Some(namespace_client_properties) = namespace_client_properties {
+            builder = builder.namespace_client_properties(namespace_client_properties);
+        }
         #[cfg(feature = "remote")]
         if let Some(client_config) = client_config {
             builder = builder.client_config(client_config.into());

python/src/expr.rs

@@ -17,7 +17,7 @@ use pyo3::{Bound, PyAny, PyResult, exceptions::PyValueError, prelude::*, pyfunct
 /// [`expr_lit`] and combined with the methods on this struct.  On the Python
 /// side a thin wrapper class (`lancedb.expr.Expr`) delegates to these methods
 /// and adds Python operator overloads.
-#[pyclass(name = "PyExpr")]
+#[pyclass(name = "PyExpr", from_py_object)]
 #[derive(Clone)]
 pub struct PyExpr(pub DfExpr);
 

python/src/header.rs

@@ -33,7 +33,7 @@ impl PyHeaderProvider {
                 Ok(headers_py) => {
                     // Convert Python dict to Rust HashMap
                     let bound_headers = headers_py.bind(py);
-                    let dict: &Bound<PyDict> = bound_headers.downcast().map_err(|e| {
+                    let dict: &Bound<PyDict> = bound_headers.cast().map_err(|e| {
                         format!("HeaderProvider.get_headers must return a dict: {}", e)
                     })?;
 

python/src/index.rs

@@ -13,7 +13,7 @@ use pyo3::{
     Bound, FromPyObject, PyAny, PyResult, Python,
     exceptions::{PyKeyError, PyValueError},
     intern, pyclass, pymethods,
-    types::PyAnyMethods,
+    types::{PyAnyMethods, PyString},
 };
 
 use crate::util::parse_distance_type;
@@ -22,7 +22,7 @@ pub fn class_name(ob: &'_ Bound<'_, PyAny>) -> PyResult<String> {
     let full_name = ob
         .getattr(intern!(ob.py(), "__class__"))?
         .getattr(intern!(ob.py(), "__name__"))?;
-    let full_name = full_name.downcast()?.to_string_lossy();
+    let full_name = full_name.cast::<PyString>()?.to_string_lossy();
 
     match full_name.rsplit_once('.') {
         Some((_, name)) => Ok(name.to_string()),

python/src/lib.rs

@@ -28,6 +28,7 @@ pub mod index;
 pub mod namespace;
 pub mod permutation;
 pub mod query;
+pub mod runtime;
 pub mod session;
 pub mod table;
 pub mod util;

python/src/namespace.rs

@@ -183,7 +183,7 @@ async fn call_py_method_primitive<Req, Resp>(
 ) -> lance_core::Result<Resp>
 where
     Req: serde::Serialize + Send + 'static,
-    Resp: for<'py> pyo3::FromPyObject<'py> + Send + 'static,
+    Resp: for<'a, 'py> pyo3::FromPyObject<'a, 'py> + Send + 'static,
 {
     let request_json = serde_json::to_string(&request).map_err(|e| {
         lance_core::Error::io(format!(
@@ -203,7 +203,7 @@ where
 
             // Call the Python method
             let result = py_namespace.call_method1(py, method_name, (request_arg,))?;
-            let value: Resp = result.extract(py)?;
+            let value: Resp = result.extract(py).map_err(Into::into)?;
             Ok::<_, PyErr>(value)
         })
     })

python/src/permutation.rs

@@ -4,7 +4,7 @@
 use std::sync::{Arc, Mutex};
 
 use crate::{
-    arrow::RecordBatchStream, connection::Connection, error::PythonErrorExt, table::Table,
+    arrow::RecordBatchStream, error::PythonErrorExt, runtime::future_into_py, table::Table,
 };
 use arrow::pyarrow::{PyArrowType, ToPyArrow};
 use lancedb::{
@@ -21,16 +21,15 @@ use pyo3::{
     pyclass, pymethods,
     types::{PyAnyMethods, PyDict, PyDictMethods, PyType},
 };
-use pyo3_async_runtimes::tokio::future_into_py;
 
 fn table_from_py<'a>(table: Bound<'a, PyAny>) -> PyResult<Bound<'a, Table>> {
     if table.hasattr("_inner")? {
-        Ok(table.getattr("_inner")?.downcast_into::<Table>()?)
+        Ok(table.getattr("_inner")?.cast_into::<Table>()?)
     } else if table.hasattr("_table")? {
         Ok(table
             .getattr("_table")?
             .getattr("_inner")?
-            .downcast_into::<Table>()?)
+            .cast_into::<Table>()?)
     } else {
         Err(PyRuntimeError::new_err(
             "Provided table does not appear to be a Table or RemoteTable instance",
@@ -80,24 +79,6 @@ impl PyAsyncPermutationBuilder {
 
 #[pymethods]
 impl PyAsyncPermutationBuilder {
-    #[pyo3(signature = (database, table_name))]
-    pub fn persist(
-        slf: PyRefMut<'_, Self>,
-        database: Bound<'_, PyAny>,
-        table_name: String,
-    ) -> PyResult<Self> {
-        let conn = if database.hasattr("_conn")? {
-            database
-                .getattr("_conn")?
-                .getattr("_inner")?
-                .downcast_into::<Connection>()?
-        } else {
-            database.getattr("_inner")?.downcast_into::<Connection>()?
-        };
-        let database = conn.borrow().database()?;
-        slf.modify(|builder| builder.persist(database, table_name))
-    }
-
     #[pyo3(signature = (*, ratios=None, counts=None, fixed=None, seed=None, split_names=None))]
     pub fn split_random(
         slf: PyRefMut<'_, Self>,
@@ -243,7 +224,7 @@ impl PyPermutationReader {
         let Some(selection) = selection else {
             return Ok(Select::All);
         };
-        let selection = selection.downcast_into::<PyDict>()?;
+        let selection = selection.cast_into::<PyDict>()?;
         let selection = selection
             .iter()
             .map(|(key, value)| {

python/src/query.rs

@@ -4,6 +4,11 @@
 use std::sync::Arc;
 use std::time::Duration;
 
+use crate::expr::PyExpr;
+use crate::runtime::future_into_py;
+use crate::util::parse_distance_type;
+use crate::{arrow::RecordBatchStream, util::PyLanceDB};
+use crate::{error::PythonErrorExt, index::class_name};
 use arrow::array::Array;
 use arrow::array::ArrayData;
 use arrow::array::make_array;
@@ -33,19 +38,16 @@ use pyo3::pyfunction;
 use pyo3::pymethods;
 use pyo3::types::PyList;
 use pyo3::types::{PyDict, PyString};
-use pyo3::{FromPyObject, exceptions::PyRuntimeError};
+use pyo3::{Borrowed, FromPyObject, exceptions::PyRuntimeError};
 use pyo3::{PyErr, pyclass};
 use pyo3::{exceptions::PyValueError, intern};
-use pyo3_async_runtimes::tokio::future_into_py;
 
-use crate::expr::PyExpr;
+impl<'a, 'py> FromPyObject<'a, 'py> for PyLanceDB<FtsQuery> {
-use crate::util::parse_distance_type;
+    type Error = PyErr;
-use crate::{arrow::RecordBatchStream, util::PyLanceDB};
-use crate::{error::PythonErrorExt, index::class_name};
 
-impl FromPyObject<'_> for PyLanceDB<FtsQuery> {
+    fn extract(ob: Borrowed<'a, 'py, PyAny>) -> PyResult<Self> {
-    fn extract_bound(ob: &Bound<'_, PyAny>) -> PyResult<Self> {
+        let ob = ob.to_owned();
-        match class_name(ob)?.as_str() {
+        match class_name(&ob)?.as_str() {
             "MatchQuery" => {
                 let query = ob.getattr("query")?.extract()?;
                 let column = ob.getattr("column")?.extract()?;
@@ -424,7 +426,7 @@ impl Query {
                 "Query text is required for nearest_to_text",
             ))?;
 
-        let query = if let Ok(query_text) = fts_query.downcast::<PyString>() {
+        let query = if let Ok(query_text) = fts_query.cast::<PyString>() {
             let mut query_text = query_text.to_string();
             let columns = query
                 .get_item("columns")?
@@ -606,7 +608,7 @@ impl TakeQuery {
     }
 }
 
-#[pyclass]
+#[pyclass(from_py_object)]
 #[derive(Clone)]
 pub struct FTSQuery {
     inner: LanceDbQuery,
@@ -735,7 +737,7 @@ impl FTSQuery {
     }
 }
 
-#[pyclass]
+#[pyclass(from_py_object)]
 #[derive(Clone)]
 pub struct VectorQuery {
     inner: LanceDbVectorQuery,

python/src/runtime.rs

@@ -0,0 +1,142 @@
+// SPDX-License-Identifier: Apache-2.0
+// SPDX-FileCopyrightText: Copyright The LanceDB Authors
+
+//! Fork-safe wrapper around tokio + pyo3-async-runtimes.
+//!
+//! `pyo3_async_runtimes::tokio` keeps its multi-threaded runtime in a
+//! `OnceLock` that can never be replaced.  Tokio's worker threads do not
+//! survive `fork()`, so once a child inherits a "frozen" runtime, every
+//! `future_into_py` call hangs forever.
+//!
+//! We sidestep the global by routing every future through our own
+//! [`LanceRuntime`] (a [`pyo3_async_runtimes::generic::Runtime`] impl) backed
+//! by an [`AtomicPtr`] to a tokio runtime that we own.  A `pthread_atfork`
+//! child handler nulls the pointer; the next `spawn` rebuilds the runtime in
+//! the child.  This mirrors the pattern used in the Lance Python bindings.
+
+use std::future::Future;
+use std::pin::Pin;
+use std::sync::atomic::{AtomicBool, AtomicPtr, Ordering};
+
+use pyo3::{Bound, PyAny, PyResult, Python, conversion::IntoPyObject};
+use pyo3_async_runtimes::{
+    TaskLocals,
+    generic::{ContextExt, JoinError, Runtime},
+};
+use tokio::{runtime, task};
+
+static RUNTIME: AtomicPtr<runtime::Runtime> = AtomicPtr::new(std::ptr::null_mut());
+static RUNTIME_INSTALLING: AtomicBool = AtomicBool::new(false);
+static ATFORK_INSTALLED: AtomicBool = AtomicBool::new(false);
+
+fn create_runtime() -> runtime::Runtime {
+    runtime::Builder::new_multi_thread()
+        .enable_all()
+        .thread_name("lancedb-tokio-worker")
+        .build()
+        .expect("Failed to build tokio runtime")
+}
+
+fn get_runtime() -> &'static runtime::Runtime {
+    loop {
+        let ptr = RUNTIME.load(Ordering::SeqCst);
+        if !ptr.is_null() {
+            return unsafe { &*ptr };
+        }
+        if !RUNTIME_INSTALLING.fetch_or(true, Ordering::SeqCst) {
+            break;
+        }
+        std::thread::yield_now();
+    }
+    if !ATFORK_INSTALLED.fetch_or(true, Ordering::SeqCst) {
+        install_atfork();
+    }
+    let new_ptr = Box::into_raw(Box::new(create_runtime()));
+    RUNTIME.store(new_ptr, Ordering::SeqCst);
+    unsafe { &*new_ptr }
+}
+
+/// Runs in async-signal context after `fork()` in the child.  We can only
+/// touch atomics here; we deliberately leak the previous runtime because
+/// dropping a tokio `Runtime` would try to join its (now-dead) worker
+/// threads and hang.
+extern "C" fn atfork_child() {
+    RUNTIME.store(std::ptr::null_mut(), Ordering::SeqCst);
+    RUNTIME_INSTALLING.store(false, Ordering::SeqCst);
+}
+
+#[cfg(not(windows))]
+fn install_atfork() {
+    unsafe { libc::pthread_atfork(None, None, Some(atfork_child)) };
+}
+
+#[cfg(windows)]
+fn install_atfork() {}
+
+/// Marker type implementing [`Runtime`] over our fork-safe runtime slot.
+pub struct LanceRuntime;
+
+/// Newtype wrapper around `tokio::task::JoinError` so we can implement the
+/// foreign [`JoinError`] trait without violating orphan rules.
+pub struct LanceJoinError(task::JoinError);
+
+impl JoinError for LanceJoinError {
+    fn is_panic(&self) -> bool {
+        self.0.is_panic()
+    }
+    fn into_panic(self) -> Box<dyn std::any::Any + Send + 'static> {
+        self.0.into_panic()
+    }
+}
+
+impl Runtime for LanceRuntime {
+    type JoinError = LanceJoinError;
+    type JoinHandle = Pin<Box<dyn Future<Output = Result<(), Self::JoinError>> + Send>>;
+
+    fn spawn<F>(fut: F) -> Self::JoinHandle
+    where
+        F: Future<Output = ()> + Send + 'static,
+    {
+        let handle = get_runtime().spawn(fut);
+        Box::pin(async move { handle.await.map_err(LanceJoinError) })
+    }
+
+    fn spawn_blocking<F>(f: F) -> Self::JoinHandle
+    where
+        F: FnOnce() + Send + 'static,
+    {
+        let handle = get_runtime().spawn_blocking(f);
+        Box::pin(async move { handle.await.map_err(LanceJoinError) })
+    }
+}
+
+tokio::task_local! {
+    static TASK_LOCALS: std::cell::OnceCell<TaskLocals>;
+}
+
+impl ContextExt for LanceRuntime {
+    fn scope<F, R>(locals: TaskLocals, fut: F) -> Pin<Box<dyn Future<Output = R> + Send>>
+    where
+        F: Future<Output = R> + Send + 'static,
+    {
+        let cell = std::cell::OnceCell::new();
+        cell.set(locals).unwrap();
+        Box::pin(TASK_LOCALS.scope(cell, fut))
+    }
+
+    fn get_task_locals() -> Option<TaskLocals> {
+        TASK_LOCALS
+            .try_with(|c| c.get().cloned())
+            .unwrap_or_default()
+    }
+}
+
+/// Drop-in replacement for `pyo3_async_runtimes::tokio::future_into_py` that
+/// uses our fork-safe runtime.
+pub fn future_into_py<F, T>(py: Python<'_>, fut: F) -> PyResult<Bound<'_, PyAny>>
+where
+    F: Future<Output = PyResult<T>> + Send + 'static,
+    T: for<'py> IntoPyObject<'py> + Send + 'static,
+{
+    pyo3_async_runtimes::generic::future_into_py::<LanceRuntime, _, T>(py, fut)
+}

python/src/session.rs

@@ -11,7 +11,7 @@ use pyo3::{PyResult, pyclass, pymethods};
 /// Sessions allow you to configure cache sizes for index and metadata caches,
 /// which can significantly impact memory use and performance. They can
 /// also be re-used across multiple connections to share the same cache state.
-#[pyclass]
+#[pyclass(from_py_object)]
 #[derive(Clone)]
 pub struct Session {
     pub(crate) inner: Arc<LanceSession>,

python/src/table.rs

@@ -2,6 +2,7 @@
 // SPDX-FileCopyrightText: Copyright The LanceDB Authors
 use std::{collections::HashMap, sync::Arc};
 
+use crate::runtime::future_into_py;
 use crate::{
     connection::Connection,
     error::PythonErrorExt,
@@ -24,12 +25,11 @@ use pyo3::{
     pyclass, pymethods,
     types::{IntoPyDict, PyAnyMethods, PyDict, PyDictMethods},
 };
-use pyo3_async_runtimes::tokio::future_into_py;
 
 mod scannable;
 
 /// Statistics about a compaction operation.
-#[pyclass(get_all)]
+#[pyclass(get_all, from_py_object)]
 #[derive(Clone, Debug)]
 pub struct CompactionStats {
     /// The number of fragments removed
@@ -43,7 +43,7 @@ pub struct CompactionStats {
 }
 
 /// Statistics about a cleanup operation
-#[pyclass(get_all)]
+#[pyclass(get_all, from_py_object)]
 #[derive(Clone, Debug)]
 pub struct RemovalStats {
     /// The number of bytes removed
@@ -53,7 +53,7 @@ pub struct RemovalStats {
 }
 
 /// Statistics about an optimize operation
-#[pyclass(get_all)]
+#[pyclass(get_all, from_py_object)]
 #[derive(Clone, Debug)]
 pub struct OptimizeStats {
     /// Statistics about the compaction operation
@@ -62,7 +62,7 @@ pub struct OptimizeStats {
     pub prune: RemovalStats,
 }
 
-#[pyclass(get_all)]
+#[pyclass(get_all, from_py_object)]
 #[derive(Clone, Debug)]
 pub struct UpdateResult {
     pub rows_updated: u64,
@@ -88,7 +88,7 @@ impl From<lancedb::table::UpdateResult> for UpdateResult {
     }
 }
 
-#[pyclass(get_all)]
+#[pyclass(get_all, from_py_object)]
 #[derive(Clone, Debug)]
 pub struct AddResult {
     pub version: u64,
@@ -109,7 +109,7 @@ impl From<lancedb::table::AddResult> for AddResult {
     }
 }
 
-#[pyclass(get_all)]
+#[pyclass(get_all, from_py_object)]
 #[derive(Clone, Debug)]
 pub struct DeleteResult {
     pub num_deleted_rows: u64,
@@ -135,7 +135,7 @@ impl From<lancedb::table::DeleteResult> for DeleteResult {
     }
 }
 
-#[pyclass(get_all)]
+#[pyclass(get_all, from_py_object)]
 #[derive(Clone, Debug)]
 pub struct MergeResult {
     pub version: u64,
@@ -171,7 +171,7 @@ impl From<lancedb::table::MergeResult> for MergeResult {
     }
 }
 
-#[pyclass(get_all)]
+#[pyclass(get_all, from_py_object)]
 #[derive(Clone, Debug)]
 pub struct AddColumnsResult {
     pub version: u64,
@@ -192,7 +192,7 @@ impl From<lancedb::table::AddColumnsResult> for AddColumnsResult {
     }
 }
 
-#[pyclass(get_all)]
+#[pyclass(get_all, from_py_object)]
 #[derive(Clone, Debug)]
 pub struct AlterColumnsResult {
     pub version: u64,
@@ -213,7 +213,7 @@ impl From<lancedb::table::AlterColumnsResult> for AlterColumnsResult {
     }
 }
 
-#[pyclass(get_all)]
+#[pyclass(get_all, from_py_object)]
 #[derive(Clone, Debug)]
 pub struct DropColumnsResult {
     pub version: u64,

python/src/table/scannable.rs

@@ -126,8 +126,11 @@ impl Scannable for PyScannable {
     }
 }
 
-impl<'py> FromPyObject<'py> for PyScannable {
+impl<'a, 'py> FromPyObject<'a, 'py> for PyScannable {
-    fn extract_bound(ob: &pyo3::Bound<'py, PyAny>) -> pyo3::PyResult<Self> {
+    type Error = pyo3::PyErr;
+
+    fn extract(ob: pyo3::Borrowed<'a, 'py, PyAny>) -> pyo3::PyResult<Self> {
+        let ob = ob.to_owned();
         // Convert from Scannable dataclass.
         let schema: PyArrowType<Schema> = ob.getattr("schema")?.extract()?;
         let schema = Arc::new(schema.0);

python/uv.lock

@@ -1996,7 +1996,6 @@ tests = [
     { name = "pytest-mock" },
     { name = "pytz" },
     { name = "requests" },
-    { name = "tantivy" },
 ]
 
 [package.metadata]
@@ -2050,7 +2049,6 @@ requires-dist = [
     { name = "sentence-transformers", marker = "extra == 'embeddings'", specifier = ">=2.2.0" },
     { name = "sentencepiece", marker = "extra == 'embeddings'", specifier = ">=0.1.99" },
     { name = "sentencepiece", marker = "extra == 'siglip'" },
-    { name = "tantivy", marker = "extra == 'tests'", specifier = ">=0.20.0" },
     { name = "torch", marker = "extra == 'clip'" },
     { name = "torch", marker = "extra == 'embeddings'", specifier = ">=2.0.0" },
     { name = "torch", marker = "extra == 'siglip'" },
@@ -4779,44 +4777,6 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/40/44/4a5f08c96eb108af5cb50b41f76142f0afa346dfa99d5296fe7202a11854/tabulate-0.9.0-py3-none-any.whl", hash = "sha256:024ca478df22e9340661486f85298cff5f6dcdba14f3813e8830015b9ed1948f", size = 35252, upload-time = "2022-10-06T17:21:44.262Z" },
 ]
 
-[[package]]
-name = "tantivy"
-version = "0.25.1"
-source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/1b/f9/0cd3955d155d3e3ef74b864769514dd191e5dacba9f0beb7af2d914942ce/tantivy-0.25.1.tar.gz", hash = "sha256:68a3314699a7d18fcf338b52bae8ce46a97dde1128a3e47e33fa4db7f71f265e", size = 75120, upload-time = "2025-12-02T11:57:12.997Z" }
-wheels = [
-    { url = "https://files.pythonhosted.org/packages/80/f7/2276bed3bed983ce2970dc70e3571f372587fe4f5f2bac1d6d617df08fa3/tantivy-0.25.1-cp310-cp310-macosx_10_12_x86_64.macosx_11_0_arm64.macosx_10_12_universal2.whl", hash = "sha256:7aa587a3dc9470584cacf5e3640fee93d12ec5f10109669c1f47c4e90820b958", size = 7638510, upload-time = "2025-12-02T11:56:08.754Z" },
-    { url = "https://files.pythonhosted.org/packages/20/8c/078dc50570e243414356b05633f52fe544b85179281ffa9f1fe05d76bbd8/tantivy-0.25.1-cp310-cp310-macosx_10_12_x86_64.whl", hash = "sha256:56d77fe667595693d9fa5f0b4545776d84da9526bab0273b3fc6c7536dc0d8a2", size = 3932659, upload-time = "2025-12-02T11:56:10.621Z" },
-    { url = "https://files.pythonhosted.org/packages/bd/dc/281c48436a1e3178b58fe463af314434fe0f3a4ec0c7588a362900e0c69e/tantivy-0.25.1-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:5ba8c347cd48595fcaeabb28a909ebce92cf9c5e5c84ab5ba1136a280a307b5c", size = 4197430, upload-time = "2025-12-02T11:56:12.65Z" },
-    { url = "https://files.pythonhosted.org/packages/7b/6c/61e6e0b0a350007d10a9b66a35703361d3345e14e7a7cc83494776b2a054/tantivy-0.25.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:aa7c4932e8fde1f09f2d46226060e827e197c2749abdc6129d73a752773adc38", size = 4184055, upload-time = "2025-12-02T11:56:14.647Z" },
-    { url = "https://files.pythonhosted.org/packages/5f/fd/0eb059b12f0b6f91623a54a46448a83b7f716d08f3bca68c095d697b85da/tantivy-0.25.1-cp310-cp310-win_amd64.whl", hash = "sha256:afcfc5dbb0bcd5d24531f4471737ae0896f33528426ab0b1dad3e427c19120f6", size = 3424134, upload-time = "2025-12-02T11:56:16.242Z" },
-    { url = "https://files.pythonhosted.org/packages/4e/7a/8a277f377e8a151fc0e71d4ffc1114aefb6e5e1c7dd609fed0955cf34ed8/tantivy-0.25.1-cp311-cp311-macosx_10_12_x86_64.macosx_11_0_arm64.macosx_10_12_universal2.whl", hash = "sha256:d363d7b4207d3a5aa7f0d212420df35bed18bdb6bae26a2a8bd57428388b7c29", size = 7637033, upload-time = "2025-12-02T11:56:18.104Z" },
-    { url = "https://files.pythonhosted.org/packages/71/31/8b4acdedfc9f9a2d04b1340d07eef5213d6f151d1e18da0cb423e5f090d2/tantivy-0.25.1-cp311-cp311-macosx_10_12_x86_64.whl", hash = "sha256:8f4389cf1d889a1df7c5a3195806b4b56c37cee10d8a26faaa0dea35a867b5ff", size = 3932180, upload-time = "2025-12-02T11:56:19.833Z" },
-    { url = "https://files.pythonhosted.org/packages/2f/dc/3e8499c21b4b9795e8f2fc54c68ce5b92905aaeadadaa56ecfa9180b11b1/tantivy-0.25.1-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:99864c09fc54652c3c2486cdf13f86cdc8200f4b481569cb291e095ca5d496e5", size = 4197620, upload-time = "2025-12-02T11:56:21.496Z" },
-    { url = "https://files.pythonhosted.org/packages/f8/8e/f2ce62fffc811eb62bead92c7b23c2e218f817cbd54c4f3b802e03ba1438/tantivy-0.25.1-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:05abf37ddbc5063c575548be0d62931629c086bff7a5a1b67cf5a8f5ebf4cd8c", size = 4183794, upload-time = "2025-12-02T11:56:23.215Z" },
-    { url = "https://files.pythonhosted.org/packages/de/64/24e2891b0ba3fd9853e10c296095a33b89bf3efd65e29da1ee5dae736040/tantivy-0.25.1-cp311-cp311-win_amd64.whl", hash = "sha256:f307ee8ad21597b0be23af83008fd66cfd5f958cdfa24ec0aaa08a38e86bbef4", size = 3424235, upload-time = "2025-12-02T11:56:25.172Z" },
-    { url = "https://files.pythonhosted.org/packages/41/e7/6849c713ed0996c7628324c60512c4882006f0a62145e56c624a93407f90/tantivy-0.25.1-cp312-cp312-macosx_10_12_x86_64.macosx_11_0_arm64.macosx_10_12_universal2.whl", hash = "sha256:90fd919e5f611809f746560ecf36eb9be824dec62e21ae17a27243759edb9aa1", size = 7621494, upload-time = "2025-12-02T11:56:27.069Z" },
-    { url = "https://files.pythonhosted.org/packages/c5/22/c3d8294600dc6e7fa350daef9ff337d3c06e132b81df727de9f7a50c692a/tantivy-0.25.1-cp312-cp312-macosx_10_12_x86_64.whl", hash = "sha256:4613c7cf6c23f3a97989819690a0f956d799354957de7a204abcc60083cebe02", size = 3925219, upload-time = "2025-12-02T11:56:29.403Z" },
-    { url = "https://files.pythonhosted.org/packages/41/fc/cbb1df71dd44c9110eff4eaaeda9d44f2d06182fe0452193be20ddfba93f/tantivy-0.25.1-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:c477bd20b4df804d57dfc5033431bef27cde605695ae141b03abbf6ebc069129", size = 4198699, upload-time = "2025-12-02T11:56:31.359Z" },
-    { url = "https://files.pythonhosted.org/packages/47/4d/71abb78b774073c3ce12a4faa4351a9d910a71ffa3659526affba163873d/tantivy-0.25.1-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:f9b1a1ba1113c523c7ff7b10f282d6c4074006f7ef8d71e1d973d51bf7291ddb", size = 4183585, upload-time = "2025-12-02T11:56:33.317Z" },
-    { url = "https://files.pythonhosted.org/packages/be/16/3f00cd7ec458b92a0e977960af9ddfbeb762127d9acc68da9094a1fda556/tantivy-0.25.1-cp312-cp312-win_amd64.whl", hash = "sha256:9de0bafd3bd7ac9f8f82d53e17562e9db11a5af308fe5185c4bd86feaddbe4a6", size = 3424622, upload-time = "2025-12-02T11:56:34.788Z" },
-    { url = "https://files.pythonhosted.org/packages/3d/25/73cfbcf1a8ea49be6c42817431cac46b70a119fe64da903fcc2d92b5b511/tantivy-0.25.1-cp313-cp313-macosx_10_12_x86_64.macosx_11_0_arm64.macosx_10_12_universal2.whl", hash = "sha256:f51ff7196c6f31719202080ed8372d5e3d51e92c749c032fb8234f012e99744c", size = 7622530, upload-time = "2025-12-02T11:56:36.839Z" },
-    { url = "https://files.pythonhosted.org/packages/12/c8/c0d7591cdf4f7e7a9fc4da786d1ca8cd1aacffaa2be16ea6d401a8e4a566/tantivy-0.25.1-cp313-cp313-macosx_10_12_x86_64.whl", hash = "sha256:550e63321bfcacc003859f2fa29c1e8e56450807b3c9a501c1add27cfb9236d9", size = 3925637, upload-time = "2025-12-02T11:56:38.425Z" },
-    { url = "https://files.pythonhosted.org/packages/3a/09/bedfc223bffec7641b417dd7ab071134b2ef8f8550e9b1fb6014657ef52e/tantivy-0.25.1-cp313-cp313-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:fde31cc8d6e122faf7902aeea32bc008a429a6e8904e34d3468126a3ec01b016", size = 4197322, upload-time = "2025-12-02T11:56:40.411Z" },
-    { url = "https://files.pythonhosted.org/packages/f5/f1/1fa5183500c8042200c9f2b840d34f5bbcfb434a1ee750e7132262d2a5c9/tantivy-0.25.1-cp313-cp313-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:b11bd5a518b0be645320b47af8493f6a40c4f3234313e37adcf4534a564d27dd", size = 4183143, upload-time = "2025-12-02T11:56:42.048Z" },
-    { url = "https://files.pythonhosted.org/packages/d5/74/a4c4f4eb95888ccb784da3b017aa0625ab1ac411bf5d022a9a797d9a2334/tantivy-0.25.1-cp313-cp313-win_amd64.whl", hash = "sha256:cc7fe88853e06b3251ee4fa42b7a2038727f850c8765bcc8167cfc73585dd24e", size = 3423491, upload-time = "2025-12-02T11:56:43.858Z" },
-    { url = "https://files.pythonhosted.org/packages/8b/2f/581519492226f97d23bd0adc95dad991ebeaa73ea6abc8bff389a3096d9a/tantivy-0.25.1-cp313-cp313t-macosx_10_12_x86_64.macosx_11_0_arm64.macosx_10_12_universal2.whl", hash = "sha256:dae99e75b7eaa9bf5bd16ab106b416370f08c135aed0e117d62a3201cd1ffe36", size = 7610316, upload-time = "2025-12-02T11:56:45.927Z" },
-    { url = "https://files.pythonhosted.org/packages/91/40/5d7bc315ab9e6a22c5572656e8ada1c836cfa96dccf533377504fbc3c9d9/tantivy-0.25.1-cp313-cp313t-macosx_10_12_x86_64.whl", hash = "sha256:506e9533c5ef4d3df43bad64ffecc0aa97c76e361ea610815dc3a20a9d6b30b3", size = 3919882, upload-time = "2025-12-02T11:56:48.469Z" },
-    { url = "https://files.pythonhosted.org/packages/02/b9/e0ef2f57a6a72444cb66c2ffbc310ab33ffaace275f1c4b0319d84ea3f18/tantivy-0.25.1-cp313-cp313t-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:5dbd4f8f264dacbcc9dee542832da2173fd53deaaea03f082d95214f8b5ed6bc", size = 4196031, upload-time = "2025-12-02T11:56:50.151Z" },
-    { url = "https://files.pythonhosted.org/packages/1e/02/bf3f8cacfd08642e14a73f7956a3fb95d58119132c98c121b9065a1f8615/tantivy-0.25.1-cp313-cp313t-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:824c643ccb640dd9e35e00c5d5054ddf3323f56fe4219d57d428a9eeea13d22c", size = 4183437, upload-time = "2025-12-02T11:56:51.818Z" },
-    { url = "https://files.pythonhosted.org/packages/9c/83/afa90e570198e2d1139dd567bec3c9cf44d8c54f63a649f16d711ede02f5/tantivy-0.25.1-cp313-cp313t-win_amd64.whl", hash = "sha256:09c987b840afcebac817836ac08407eff17272d8aa60ce6e291f89c81830221d", size = 3419409, upload-time = "2025-12-02T11:56:53.451Z" },
-    { url = "https://files.pythonhosted.org/packages/ff/44/9f1d67aa5030f7eebc966c863d1316a510a971dd8bb45651df4acdfae9ed/tantivy-0.25.1-cp314-cp314-macosx_10_12_x86_64.macosx_11_0_arm64.macosx_10_12_universal2.whl", hash = "sha256:7f5d29ae85dd0f23df8d15b3e7b341d4f9eb5a446bbb9640df48ac1f6d9e0c6c", size = 7623723, upload-time = "2025-12-02T11:56:55.066Z" },
-    { url = "https://files.pythonhosted.org/packages/db/30/6e085bd3ed9d12da3c91c185854abd70f9dfd35fb36a75ea98428d42c30b/tantivy-0.25.1-cp314-cp314-macosx_10_12_x86_64.whl", hash = "sha256:f2d2938fb69a74fc1bb36edfaf7f0d1596fa1264db0f377bda2195c58bcb6245", size = 3926243, upload-time = "2025-12-02T11:56:57.058Z" },
-    { url = "https://files.pythonhosted.org/packages/32/f5/a00d65433430f51718e5cc6938df571765d7c4e03aedec5aef4ab567aa9b/tantivy-0.25.1-cp314-cp314-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:4f5ff124c4802558e627091e780b362ca944169736caba5a372eef39a79d0ae0", size = 4207186, upload-time = "2025-12-02T11:56:58.803Z" },
-    { url = "https://files.pythonhosted.org/packages/19/63/61bdb12fc95f2a7f77bd419a5149bfa9f28caa76cb569bf2b6b06e1d033e/tantivy-0.25.1-cp314-cp314-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:43b80ef62a340416139c93d19264e5f808da48e04f9305f1092b8ed22be0a5be", size = 4187312, upload-time = "2025-12-02T11:57:00.595Z" },
-    { url = "https://files.pythonhosted.org/packages/b7/de/e39c0b01d59019bf5c38face8b81defbc4a68cebf5e0c53bcb2cd715a449/tantivy-0.25.1-cp314-cp314-win_amd64.whl", hash = "sha256:286b654f40c70c1e6b64b9bc7031ed0bf5c440f5bffeaeeee21a0ee6cc39f0e2", size = 3436535, upload-time = "2025-12-02T11:57:02.267Z" },
-]
-
 [[package]]
 name = "threadpoolctl"
 version = "3.6.0"

rust/lancedb/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "lancedb"
-version = "0.28.0-beta.7"
+version = "0.28.0-beta.11"
 edition.workspace = true
 description = "LanceDB: A serverless, low-latency vector database for AI applications"
 license.workspace = true
@@ -108,10 +108,20 @@ test-log = "0.2"
 
 [features]
 default = []
-aws = ["lance/aws", "lance-io/aws", "lance-namespace-impls/dir-aws"]
+aws = [
+    "lance/aws",
+    "lance-io/aws",
+    "lance-namespace-impls/dir-aws",
+    "object_store/aws",
+]
 oss = ["lance/oss", "lance-io/oss", "lance-namespace-impls/dir-oss"]
 gcs = ["lance/gcp", "lance-io/gcp", "lance-namespace-impls/dir-gcp"]
-azure = ["lance/azure", "lance-io/azure", "lance-namespace-impls/dir-azure"]
+azure = [
+    "lance/azure",
+    "lance-io/azure",
+    "lance-namespace-impls/dir-azure",
+    "lance-namespace-impls/credential-vendor-azure",
+]
 huggingface = [
     "lance/huggingface",
     "lance-io/huggingface",

rust/lancedb/src/connection.rs

@@ -590,6 +590,15 @@ pub struct ConnectRequest {
     /// storage options.
     pub namespace_client_properties: HashMap<String, String>,
 
+    /// Use directory namespace manifests as the source of truth for native
+    /// LanceDB table metadata.
+    ///
+    /// When enabled for a local/native connection, LanceDB returns a
+    /// namespace-backed database directly. Directory listing fallback remains
+    /// enabled for migration, and directory-listing-to-manifest migration is
+    /// forced on.
+    pub manifest_enabled: bool,
+
     /// The interval at which to check for updates from other processes.
     ///
     /// If None, then consistency is not checked. For performance
@@ -630,6 +639,7 @@ impl ConnectBuilder {
                 read_consistency_interval: None,
                 options: HashMap::new(),
                 namespace_client_properties: HashMap::new(),
+                manifest_enabled: false,
                 session: None,
             },
             embedding_registry: None,
@@ -791,6 +801,17 @@ impl ConnectBuilder {
         self
     }
 
+    /// Enable or disable manifest-backed directory namespace mode for local
+    /// native connections.
+    ///
+    /// When enabled, the connection uses the directory namespace database
+    /// directly for all table operations and forces
+    /// `dir_listing_to_manifest_migration_enabled=true`.
+    pub fn manifest_enabled(mut self, enabled: bool) -> Self {
+        self.request.manifest_enabled = enabled;
+        self
+    }
+
     /// The interval at which to check for updates from other processes. This
     /// only affects LanceDB OSS.
     ///
@@ -886,6 +907,16 @@ impl ConnectBuilder {
     pub async fn execute(self) -> Result<Connection> {
         if self.request.uri.starts_with("db") {
             self.execute_remote()
+        } else if self.request.manifest_enabled {
+            let internal = Arc::new(
+                ListingDatabase::connect_manifest_enabled_namespace_database(&self.request).await?,
+            );
+            Ok(Connection {
+                internal,
+                embedding_registry: self
+                    .embedding_registry
+                    .unwrap_or_else(|| Arc::new(MemoryRegistry::new())),
+            })
         } else {
             let internal = Arc::new(ListingDatabase::connect_with_options(&self.request).await?);
             Ok(Connection {
@@ -1132,6 +1163,9 @@ mod tests {
     use lance_testing::datagen::{BatchGenerator, IncrementingInt32};
     use tempfile::tempdir;
 
+    use crate::database::listing::{ListingDatabaseOptions, OPT_NEW_TABLE_V2_MANIFEST_PATHS};
+    use crate::database::namespace::LanceNamespaceDatabase;
+    use crate::table::NativeTable;
     use crate::test_utils::connection::new_test_connection;
 
     use super::*;
@@ -1204,6 +1238,147 @@ mod tests {
         );
     }
 
+    #[tokio::test]
+    async fn test_connect_with_manifest_enabled_uses_directory_namespace() {
+        let tmp_dir = tempdir().unwrap();
+        let uri = tmp_dir.path().to_str().unwrap();
+
+        let db = connect(uri)
+            .manifest_enabled(true)
+            .storage_option("timeout", "30s")
+            .namespace_client_property("manifest_enabled", "false")
+            .namespace_client_property("dir_listing_to_manifest_migration_enabled", "false")
+            .execute()
+            .await
+            .unwrap();
+
+        assert!(
+            db.database()
+                .as_any()
+                .downcast_ref::<LanceNamespaceDatabase>()
+                .is_some()
+        );
+        assert_eq!(db.uri(), uri);
+
+        let (ns_impl, properties) = db.namespace_client_config().await.unwrap();
+        assert_eq!(ns_impl, "dir");
+        assert_eq!(properties.get("root"), Some(&uri.to_string()));
+        assert_eq!(
+            properties.get("manifest_enabled"),
+            Some(&"true".to_string())
+        );
+        assert_eq!(
+            properties.get("dir_listing_to_manifest_migration_enabled"),
+            Some(&"true".to_string())
+        );
+        assert_eq!(properties.get("storage.timeout"), Some(&"30s".to_string()));
+    }
+
+    #[tokio::test]
+    async fn test_manifest_enabled_rejects_commit_engine_uri() {
+        let Err(err) = connect("s3+ddb://bucket/db?ddbTableName=manifest")
+            .manifest_enabled(true)
+            .execute()
+            .await
+        else {
+            panic!("expected manifest-enabled s3+ddb connection to fail");
+        };
+        assert!(
+            matches!(err, Error::NotSupported { message } if message.contains("commit engine URI schemes"))
+        );
+
+        let Err(err) = connect("s3://bucket/db?engine=ddb&ddbTableName=manifest")
+            .manifest_enabled(true)
+            .execute()
+            .await
+        else {
+            panic!("expected manifest-enabled engine query connection to fail");
+        };
+        assert!(
+            matches!(err, Error::NotSupported { message } if message.contains("commit engine"))
+        );
+    }
+
+    #[tokio::test]
+    async fn test_manifest_enabled_connection_migrates_root_listing_table() {
+        let tmp_dir = tempdir().unwrap();
+        let uri = tmp_dir.path().to_str().unwrap();
+        let schema = Arc::new(Schema::new(vec![Field::new("x", DataType::Int32, false)]));
+
+        connect(uri)
+            .execute()
+            .await
+            .unwrap()
+            .create_empty_table("legacy", schema)
+            .execute()
+            .await
+            .unwrap();
+
+        let db = connect(uri).manifest_enabled(true).execute().await.unwrap();
+        let tables = db.table_names().execute().await.unwrap();
+        assert_eq!(tables, vec!["legacy".to_string()]);
+        db.open_table("legacy").execute().await.unwrap();
+    }
+
+    #[tokio::test]
+    async fn test_manifest_enabled_preserves_new_table_options() {
+        let tmp_dir = tempdir().unwrap();
+        let uri = tmp_dir.path().to_str().unwrap();
+        let options = ListingDatabaseOptions::builder()
+            .enable_v2_manifest_paths(true)
+            .build();
+        let schema = Arc::new(Schema::new(vec![Field::new("x", DataType::Int32, false)]));
+
+        let table = connect(uri)
+            .manifest_enabled(true)
+            .database_options(&options)
+            .execute()
+            .await
+            .unwrap()
+            .create_empty_table("v1_manifest", schema)
+            .storage_option(OPT_NEW_TABLE_V2_MANIFEST_PATHS, "false")
+            .execute()
+            .await
+            .unwrap();
+
+        let native_table = table
+            .base_table()
+            .as_any()
+            .downcast_ref::<NativeTable>()
+            .unwrap();
+        assert!(!native_table.uses_v2_manifest_paths().await.unwrap());
+    }
+
+    #[tokio::test]
+    async fn test_manifest_enabled_vend_input_storage_options() {
+        let tmp_dir = tempdir().unwrap();
+        let uri = tmp_dir.path().to_str().unwrap();
+        let schema = Arc::new(Schema::new(vec![Field::new("x", DataType::Int32, false)]));
+
+        let table = connect(uri)
+            .manifest_enabled(true)
+            .storage_option("test_storage_option", "test_value")
+            .namespace_client_property("vend_input_storage_options", "true")
+            .namespace_client_property(
+                "vend_input_storage_options_refresh_interval_millis",
+                "60000",
+            )
+            .execute()
+            .await
+            .unwrap()
+            .create_empty_table("vended", schema)
+            .execute()
+            .await
+            .unwrap();
+
+        let storage_options = table.latest_storage_options().await.unwrap().unwrap();
+        assert_eq!(
+            storage_options.get("test_storage_option"),
+            Some(&"test_value".to_string())
+        );
+        assert!(storage_options.contains_key("expires_at_millis"));
+    }
+
     #[tokio::test]
     async fn test_table_names() {
         let tc = new_test_connection().await.unwrap();

rust/lancedb/src/database/listing.rs

@@ -285,7 +285,7 @@ const MIRRORED_STORE: &str = "mirroredStore";
 
 /// A connection to LanceDB
 impl ListingDatabase {
-    fn build_namespace_client_properties(
+    pub(crate) fn build_namespace_client_properties(
         uri: &str,
         storage_options: &HashMap<String, String>,
         namespace_client_properties: HashMap<String, String>,
@@ -298,6 +298,24 @@ impl ListingDatabase {
         properties
     }
 
+    pub(crate) fn build_manifest_enabled_namespace_client_properties(
+        uri: &str,
+        storage_options: &HashMap<String, String>,
+        namespace_client_properties: HashMap<String, String>,
+    ) -> HashMap<String, String> {
+        let mut properties = Self::build_namespace_client_properties(
+            uri,
+            storage_options,
+            namespace_client_properties,
+        );
+        properties.insert("manifest_enabled".to_string(), "true".to_string());
+        properties.insert(
+            "dir_listing_to_manifest_migration_enabled".to_string(),
+            "true".to_string(),
+        );
+        properties
+    }
+
     async fn connect_namespace_database(
         uri: &str,
         storage_options: HashMap<String, String>,
@@ -323,6 +341,119 @@ impl ListingDatabase {
         ))
     }
 
+    async fn prepare_namespace_root(
+        uri: &str,
+        storage_options: &HashMap<String, String>,
+        session: Arc<lance::session::Session>,
+    ) -> Result<String> {
+        match url::Url::parse(uri) {
+            Ok(url) if url.scheme().len() == 1 && cfg!(windows) => {
+                let (object_store, _) = ObjectStore::from_uri_and_params(
+                    session.store_registry(),
+                    uri,
+                    &ObjectStoreParams::default(),
+                )
+                .await?;
+                if object_store.is_local() {
+                    Self::try_create_dir(uri).context(CreateDirSnafu { path: uri })?;
+                }
+                Ok(uri.to_string())
+            }
+            Ok(mut url) => {
+                if url.scheme().contains('+') {
+                    return Err(Error::NotSupported {
+                        message: "commit engine URI schemes are not supported for manifest-enabled namespace connections".to_string(),
+                    });
+                }
+
+                for (key, value) in url.query_pairs() {
+                    if key == ENGINE {
+                        return Err(Error::NotSupported {
+                            message: format!(
+                                "commit engine '{}' is not supported for manifest-enabled namespace connections",
+                                value
+                            ),
+                        });
+                    } else if key == MIRRORED_STORE {
+                        return Err(Error::NotSupported {
+                            message: "mirrored store is not supported for manifest-enabled namespace connections"
+                                .to_string(),
+                        });
+                    }
+                }
+
+                url.set_query(None);
+                let plain_uri = url.to_string();
+
+                let os_params = ObjectStoreParams {
+                    storage_options_accessor: if storage_options.is_empty() {
+                        None
+                    } else {
+                        Some(Arc::new(StorageOptionsAccessor::with_static_options(
+                            storage_options.clone(),
+                        )))
+                    },
+                    ..Default::default()
+                };
+                let (object_store, _) = ObjectStore::from_uri_and_params(
+                    session.store_registry(),
+                    &plain_uri,
+                    &os_params,
+                )
+                .await?;
+                if object_store.is_local() {
+                    Self::try_create_dir(&plain_uri).context(CreateDirSnafu {
+                        path: plain_uri.clone(),
+                    })?;
+                }
+
+                Ok(plain_uri)
+            }
+            Err(_) => {
+                let (object_store, _) = ObjectStore::from_uri_and_params(
+                    session.store_registry(),
+                    uri,
+                    &ObjectStoreParams::default(),
+                )
+                .await?;
+                if object_store.is_local() {
+                    Self::try_create_dir(uri).context(CreateDirSnafu { path: uri })?;
+                }
+                Ok(uri.to_string())
+            }
+        }
+    }
+
+    pub(crate) async fn connect_manifest_enabled_namespace_database(
+        request: &ConnectRequest,
+    ) -> Result<LanceNamespaceDatabase> {
+        let options = ListingDatabaseOptions::parse_from_map(&request.options)?;
+        let session = request
+            .session
+            .clone()
+            .unwrap_or_else(|| Arc::new(lance::session::Session::default()));
+        let namespace_root =
+            Self::prepare_namespace_root(&request.uri, &options.storage_options, session.clone())
+                .await?;
+        let ns_properties = Self::build_manifest_enabled_namespace_client_properties(
+            &namespace_root,
+            &options.storage_options,
+            request.namespace_client_properties.clone(),
+        );
+
+        LanceNamespaceDatabase::connect_with_new_table_config(
+            "dir",
+            ns_properties,
+            options.storage_options,
+            request.read_consistency_interval,
+            Some(session),
+            HashSet::new(),
+            options.new_table_config,
+        )
+        .await
+        .map(|db| db.with_uri(request.uri.clone()))
+    }
+
     /// Connect to a listing database
     ///
     /// The URI should be a path to a directory where the tables are stored.
@@ -584,7 +715,7 @@ impl ListingDatabase {
         let commit_handler = commit_handler_from_url(&uri, &Some(object_store_params)).await?;
         for name in names {
             let dir_name = format!("{}.{}", name, LANCE_EXTENSION);
-            let full_path = self.base_path.child(dir_name.clone());
+            let full_path = self.base_path.clone().join(dir_name.clone());
 
             commit_handler.delete(&full_path).await?;
 
@@ -690,15 +821,12 @@ impl ListingDatabase {
             store_params.storage_options_accessor = Some(Arc::new(accessor));
         }
 
-        write_params.data_storage_version = self
+        write_params.data_storage_version = storage_version_override
-            .new_table_config
+            .or(write_params.data_storage_version)
-            .data_storage_version
+            .or(self.new_table_config.data_storage_version);
-            .or(storage_version_override);
 
-        if let Some(enable_v2_manifest_paths) = self
+        if let Some(enable_v2_manifest_paths) =
-            .new_table_config
+            v2_manifest_override.or(self.new_table_config.enable_v2_manifest_paths)
-            .enable_v2_manifest_paths
-            .or(v2_manifest_override)
         {
             write_params.enable_v2_manifest_paths = enable_v2_manifest_paths;
         }
@@ -1158,6 +1286,7 @@ mod tests {
             client_config: Default::default(),
             options: Default::default(),
             namespace_client_properties: Default::default(),
+            manifest_enabled: false,
             read_consistency_interval: None,
             session: None,
         };
@@ -1292,6 +1421,7 @@ mod tests {
             client_config: Default::default(),
             options: options.clone(),
             namespace_client_properties: Default::default(),
+            manifest_enabled: false,
             read_consistency_interval: None,
             session: None,
         };
@@ -1827,6 +1957,7 @@ mod tests {
             client_config: Default::default(),
             options,
             namespace_client_properties: Default::default(),
+            manifest_enabled: false,
             read_consistency_interval: None,
             session: None,
         };
@@ -1933,6 +2064,7 @@ mod tests {
             client_config: Default::default(),
             options,
             namespace_client_properties: Default::default(),
+            manifest_enabled: false,
             read_consistency_interval: None,
             session: None,
         };
@@ -2005,6 +2137,7 @@ mod tests {
             client_config: Default::default(),
             options,
             namespace_client_properties: Default::default(),
+            manifest_enabled: false,
             read_consistency_interval: None,
             session: None,
         };
@@ -2202,6 +2335,7 @@ mod tests {
             client_config: Default::default(),
             options: Default::default(),
             namespace_client_properties,
+            manifest_enabled: false,
             read_consistency_interval: None,
             session: None,
         };

rust/lancedb/src/database/namespace.rs

@@ -24,8 +24,13 @@ use lance_table::io::commit::external_manifest::ExternalManifestCommitHandler;
 
 use crate::connection::NamespaceClientPushdownOperation;
 use crate::database::ReadConsistency;
+use crate::database::listing::{
+    NewTableConfig, OPT_NEW_TABLE_ENABLE_STABLE_ROW_IDS, OPT_NEW_TABLE_STORAGE_VERSION,
+    OPT_NEW_TABLE_V2_MANIFEST_PATHS,
+};
 use crate::error::{Error, Result};
 use crate::table::NativeTable;
+use lance::dataset::WriteMode;
 
 use super::{
     BaseTable, CloneTableRequest, CreateTableMode, CreateTableRequest as DbCreateTableRequest,
@@ -49,6 +54,8 @@ pub struct LanceNamespaceDatabase {
     ns_impl: String,
     // Namespace properties used to construct the namespace client
     ns_properties: HashMap<String, String>,
+    // Options for tables created by this connection
+    new_table_config: NewTableConfig,
 }
 
 impl LanceNamespaceDatabase {
@@ -70,9 +77,15 @@ impl LanceNamespaceDatabase {
             pushdown_operations: namespace_client_pushdown_operations,
             ns_impl: namespace_client_impl,
             ns_properties: namespace_client_properties,
+            new_table_config: NewTableConfig::default(),
         }
     }
 
+    pub(crate) fn with_uri(mut self, uri: impl Into<String>) -> Self {
+        self.uri = uri.into();
+        self
+    }
+
     pub async fn connect(
         ns_impl: &str,
         ns_properties: HashMap<String, String>,
@@ -80,6 +93,27 @@ impl LanceNamespaceDatabase {
         read_consistency_interval: Option<std::time::Duration>,
         session: Option<Arc<lance::session::Session>>,
         pushdown_operations: HashSet<NamespaceClientPushdownOperation>,
+    ) -> Result<Self> {
+        Self::connect_with_new_table_config(
+            ns_impl,
+            ns_properties,
+            storage_options,
+            read_consistency_interval,
+            session,
+            pushdown_operations,
+            NewTableConfig::default(),
+        )
+        .await
+    }
+
+    pub(crate) async fn connect_with_new_table_config(
+        ns_impl: &str,
+        ns_properties: HashMap<String, String>,
+        storage_options: HashMap<String, String>,
+        read_consistency_interval: Option<std::time::Duration>,
+        session: Option<Arc<lance::session::Session>>,
+        pushdown_operations: HashSet<NamespaceClientPushdownOperation>,
+        new_table_config: NewTableConfig,
     ) -> Result<Self> {
         let mut builder = ConnectBuilder::new(ns_impl);
         for (key, value) in ns_properties.clone() {
@@ -101,8 +135,79 @@ impl LanceNamespaceDatabase {
             pushdown_operations,
             ns_impl: ns_impl.to_string(),
             ns_properties,
+            new_table_config,
         })
     }
+
+    fn extract_storage_overrides(
+        &self,
+        request: &DbCreateTableRequest,
+    ) -> Result<(
+        Option<lance_encoding::version::LanceFileVersion>,
+        Option<bool>,
+        Option<bool>,
+    )> {
+        let storage_options = request
+            .write_options
+            .lance_write_params
+            .as_ref()
+            .and_then(|p| p.store_params.as_ref())
+            .and_then(|sp| sp.storage_options());
+
+        let storage_version_override = storage_options
+            .and_then(|opts| opts.get(OPT_NEW_TABLE_STORAGE_VERSION))
+            .map(|s| s.parse::<lance_encoding::version::LanceFileVersion>())
+            .transpose()?;
+
+        let v2_manifest_override = storage_options
+            .and_then(|opts| opts.get(OPT_NEW_TABLE_V2_MANIFEST_PATHS))
+            .map(|s| s.parse::<bool>())
+            .transpose()
+            .map_err(|_| Error::InvalidInput {
+                message: "enable_v2_manifest_paths must be a boolean".to_string(),
+            })?;
+
+        let stable_row_ids_override = storage_options
+            .and_then(|opts| opts.get(OPT_NEW_TABLE_ENABLE_STABLE_ROW_IDS))
+            .map(|s| s.parse::<bool>())
+            .transpose()
+            .map_err(|_| Error::InvalidInput {
+                message: "enable_stable_row_ids must be a boolean".to_string(),
+            })?;
+
+        Ok((
+            storage_version_override,
+            v2_manifest_override,
+            stable_row_ids_override,
+        ))
+    }
+
+    fn apply_new_table_config(
+        &self,
+        params: &mut lance::dataset::WriteParams,
+        request: &DbCreateTableRequest,
+    ) -> Result<()> {
+        let (storage_version_override, v2_manifest_override, stable_row_ids_override) =
+            self.extract_storage_overrides(request)?;
+
+        params.data_storage_version = storage_version_override
+            .or(params.data_storage_version)
+            .or(self.new_table_config.data_storage_version);
+
+        if let Some(enable_v2_manifest_paths) =
+            v2_manifest_override.or(self.new_table_config.enable_v2_manifest_paths)
+        {
+            params.enable_v2_manifest_paths = enable_v2_manifest_paths;
+        }
+
+        if let Some(enable_stable_row_ids) =
+            stable_row_ids_override.or(self.new_table_config.enable_stable_row_ids)
+        {
+            params.enable_stable_row_ids = enable_stable_row_ids;
+        }
+
+        Ok(())
+    }
 }
 
 impl std::fmt::Debug for LanceNamespaceDatabase {
@@ -184,6 +289,7 @@ impl Database for LanceNamespaceDatabase {
     async fn create_table(&self, request: DbCreateTableRequest) -> Result<Arc<dyn BaseTable>> {
         let mut table_id = request.namespace_path.clone();
         table_id.push(request.name.clone());
+        let mut existing_table = None;
 
         match request.mode {
             CreateTableMode::Create => {}
@@ -192,20 +298,7 @@ impl Database for LanceNamespaceDatabase {
                     id: Some(table_id.clone()),
                     ..Default::default()
                 };
-                let describe_result = self.namespace.describe_table(describe_request).await;
+                existing_table = self.namespace.describe_table(describe_request).await.ok();
-                if describe_result.is_ok() {
-                    // Drop the existing table - must succeed
-                    let drop_request = DropTableRequest {
-                        id: Some(table_id.clone()),
-                        ..Default::default()
-                    };
-                    self.namespace
-                        .drop_table(drop_request)
-                        .await
-                        .map_err(|e| Error::Runtime {
-                            message: format!("Failed to drop existing table for overwrite: {}", e),
-                        })?;
-                }
             }
             CreateTableMode::ExistOk(_) => {
                 let describe_request = DescribeTableRequest {
@@ -240,39 +333,86 @@ impl Database for LanceNamespaceDatabase {
         };
 
         let (location, initial_storage_options, managed_versioning) = {
-            let response = self
+            if let Some(response) = existing_table {
-                .namespace
+                let loc = response.location.ok_or_else(|| Error::Runtime {
-                .declare_table(declare_request)
+                    message: "Table location is missing from describe_table response".to_string(),
-                .await
-                .map_err(|e| {
-                    let err_str = e.to_string();
-                    if matches!(request.mode, CreateTableMode::Create)
-                        && (err_str.contains("already exists")
-                            || err_str.contains("TableAlreadyExists")
-                            || err_str.contains("table already exists"))
-                    {
-                        Error::TableAlreadyExists {
-                            name: request.name.clone(),
-                        }
-                    } else {
-                        Error::Runtime {
-                            message: format!("Failed to declare table: {}", e),
-                        }
-                    }
                 })?;
-            let loc = response.location.ok_or_else(|| Error::Runtime {
+                let opts = response
-                message: "Table location is missing from declare_table response".to_string(),
+                    .storage_options
-            })?;
+                    .or_else(|| Some(self.storage_options.clone()))
-            // Use storage options from response, fall back to self.storage_options
+                    .filter(|o| !o.is_empty());
-            let opts = response
+                (loc, opts, response.managed_versioning)
-                .storage_options
+            } else {
-                .or_else(|| Some(self.storage_options.clone()))
+                match self.namespace.declare_table(declare_request).await {
-                .filter(|o| !o.is_empty());
+                    Ok(response) => {
-            (loc, opts, response.managed_versioning)
+                        let loc = response.location.ok_or_else(|| Error::Runtime {
+                            message: "Table location is missing from declare_table response"
+                                .to_string(),
+                        })?;
+                        let opts = response
+                            .storage_options
+                            .or_else(|| Some(self.storage_options.clone()))
+                            .filter(|o: &HashMap<String, String>| !o.is_empty());
+                        (loc, opts, response.managed_versioning)
+                    }
+                    Err(e)
+                        if matches!(request.mode, CreateTableMode::Create) && {
+                            let err_str = e.to_string();
+                            err_str.contains("already exists")
+                                || err_str.contains("TableAlreadyExists")
+                                || err_str.contains("table already exists")
+                        } =>
+                    {
+                        let response = self
+                            .namespace
+                            .describe_table(DescribeTableRequest {
+                                id: Some(table_id.clone()),
+                                ..Default::default()
+                            })
+                            .await
+                            .map_err(|describe_err| Error::Runtime {
+                                message: format!(
+                                    "Failed to describe existing declared table after declare conflict: {}",
+                                    describe_err
+                                ),
+                            })?;
+
+                        if response.version.is_some() && response.schema.is_some() {
+                            return Err(Error::TableAlreadyExists {
+                                name: request.name.clone(),
+                            });
+                        }
+
+                        let loc = response.location.ok_or_else(|| Error::Runtime {
+                            message: "Table location is missing from describe_table response"
+                                .to_string(),
+                        })?;
+                        let opts = response
+                            .storage_options
+                            .or_else(|| Some(self.storage_options.clone()))
+                            .filter(|o: &HashMap<String, String>| !o.is_empty());
+                        (loc, opts, response.managed_versioning)
+                    }
+                    Err(e) => {
+                        return Err(Error::Runtime {
+                            message: format!("Failed to declare table: {}", e),
+                        });
+                    }
+                }
+            }
         };
 
         // Build write params with storage options and commit handler
-        let mut params = request.write_options.lance_write_params.unwrap_or_default();
+        let mut params = request
+            .write_options
+            .lance_write_params
+            .clone()
+            .unwrap_or_default();
+        self.apply_new_table_config(&mut params, &request)?;
+
+        if matches!(request.mode, CreateTableMode::Overwrite) {
+            params.mode = WriteMode::Overwrite;
+        }
 
         // Set up storage options if provided
         if let Some(storage_opts) = initial_storage_options {
@@ -730,6 +870,58 @@ mod tests {
         assert_eq!(id_col.value(2), 30);
     }
 
+    #[tokio::test]
+    async fn test_namespace_create_table_after_declare_conflict() {
+        let tmp_dir = tempdir().unwrap();
+        let root_path = tmp_dir.path().to_str().unwrap().to_string();
+
+        let mut properties = HashMap::new();
+        properties.insert("root".to_string(), root_path);
+
+        let conn = connect_namespace("dir", properties)
+            .execute()
+            .await
+            .expect("Failed to connect to namespace");
+
+        conn.create_namespace(CreateNamespaceRequest {
+            id: Some(vec!["test_ns".into()]),
+            ..Default::default()
+        })
+        .await
+        .expect("Failed to create namespace");
+
+        let namespace_client = conn.namespace_client().await.unwrap();
+        namespace_client
+            .declare_table(DeclareTableRequest {
+                id: Some(vec!["test_ns".into(), "declared_test".into()]),
+                ..Default::default()
+            })
+            .await
+            .expect("Failed to declare table");
+
+        let test_data = create_test_data();
+        let table = conn
+            .create_table("declared_test", test_data)
+            .namespace(vec!["test_ns".into()])
+            .execute()
+            .await
+            .expect("Failed to create table after declare conflict");
+
+        let results = table
+            .query()
+            .execute()
+            .await
+            .expect("Failed to query table")
+            .try_collect::<Vec<_>>()
+            .await
+            .expect("Failed to collect results");
+
+        assert_eq!(results.len(), 1);
+        assert_eq!(results[0].num_rows(), 5);
+        assert_eq!(table.namespace(), &["test_ns"]);
+        assert_eq!(table.id(), "test_ns$declared_test");
+    }
+
     #[tokio::test]
     async fn test_namespace_create_table_exist_ok_mode() {
         // Setup: Create a temporary directory for the namespace

rust/lancedb/src/io/object_store.rs

@@ -5,11 +5,12 @@
 
 use std::{fmt::Formatter, sync::Arc};
 
-use futures::{TryFutureExt, stream::BoxStream};
+use futures::{StreamExt, TryFutureExt, stream::BoxStream};
 use lance::io::WrappingObjectStore;
 use object_store::{
-    Error, GetOptions, GetResult, ListResult, MultipartUpload, ObjectMeta, ObjectStore,
+    CopyOptions, Error, GetOptions, GetResult, ListResult, MultipartUpload, ObjectMeta,
-    PutMultipartOptions, PutOptions, PutPayload, PutResult, Result, UploadPart, path::Path,
+    ObjectStore, ObjectStoreExt, PutMultipartOptions, PutOptions, PutPayload, PutResult, Result,
+    UploadPart, path::Path,
 };
 
 use async_trait::async_trait;
@@ -93,20 +94,6 @@ impl ObjectStore for MirroringObjectStore {
         self.primary.get_opts(location, options).await
     }
 
-    async fn head(&self, location: &Path) -> Result<ObjectMeta> {
-        self.primary.head(location).await
-    }
-
-    async fn delete(&self, location: &Path) -> Result<()> {
-        if !location.primary_only() {
-            match self.secondary.delete(location).await {
-                Err(Error::NotFound { .. }) | Ok(_) => {}
-                Err(e) => return Err(e),
-            }
-        }
-        self.primary.delete(location).await
-    }
-
     fn list(&self, prefix: Option<&Path>) -> BoxStream<'static, Result<ObjectMeta>> {
         self.primary.list(prefix)
     }
@@ -115,21 +102,40 @@ impl ObjectStore for MirroringObjectStore {
         self.primary.list_with_delimiter(prefix).await
     }
 
-    async fn copy(&self, from: &Path, to: &Path) -> Result<()> {
+    fn delete_stream(
-        if to.primary_only() {
+        &self,
-            self.primary.copy(from, to).await
+        locations: BoxStream<'static, Result<Path>>,
-        } else {
+    ) -> BoxStream<'static, Result<Path>> {
-            self.secondary.copy(from, to).await?;
+        let primary = self.primary.clone();
-            self.primary.copy(from, to).await?;
+        let secondary = self.secondary.clone();
-            Ok(())
+        locations
-        }
+            .map(move |location| {
+                let primary = primary.clone();
+                let secondary = secondary.clone();
+                async move {
+                    let location = location?;
+                    if !location.primary_only() {
+                        match secondary.delete(&location).await {
+                            Err(Error::NotFound { .. }) | Ok(_) => {}
+                            Err(e) => return Err(e),
+                        }
+                    }
+                    primary.delete(&location).await?;
+                    Ok(location)
+                }
+            })
+            .buffered(10)
+            .boxed()
     }
 
-    async fn copy_if_not_exists(&self, from: &Path, to: &Path) -> Result<()> {
+    async fn copy_opts(&self, from: &Path, to: &Path, options: CopyOptions) -> Result<()> {
-        if !to.primary_only() {
+        if to.primary_only() {
-            self.secondary.copy(from, to).await?;
+            self.primary.copy_opts(from, to, options).await
+        } else {
+            self.secondary.copy_opts(from, to, options.clone()).await?;
+            self.primary.copy_opts(from, to, options).await?;
+            Ok(())
         }
-        self.primary.copy_if_not_exists(from, to).await
     }
 }
 

rust/lancedb/src/io/object_store/io_tracking.rs

@@ -10,9 +10,9 @@ use bytes::Bytes;
 use futures::stream::BoxStream;
 use lance::io::WrappingObjectStore;
 use object_store::{
-    GetOptions, GetResult, ListResult, MultipartUpload, ObjectMeta, ObjectStore,
+    CopyOptions, GetOptions, GetResult, ListResult, MultipartUpload, ObjectMeta, ObjectStore,
-    PutMultipartOptions, PutOptions, PutPayload, PutResult, Result as OSResult, UploadPart,
+    PutMultipartOptions, PutOptions, PutPayload, PutResult, RenameOptions, Result as OSResult,
-    path::Path,
+    UploadPart, path::Path,
 };
 
 #[derive(Debug, Default)]
@@ -81,11 +81,6 @@ impl IoTrackingStore {
 #[async_trait::async_trait]
 #[deny(clippy::missing_trait_methods)]
 impl ObjectStore for IoTrackingStore {
-    async fn put(&self, location: &Path, bytes: PutPayload) -> OSResult<PutResult> {
-        self.record_write(bytes.content_length() as u64);
-        self.target.put(location, bytes).await
-    }
-
     async fn put_opts(
         &self,
         location: &Path,
@@ -96,14 +91,6 @@ impl ObjectStore for IoTrackingStore {
         self.target.put_opts(location, bytes, opts).await
     }
 
-    async fn put_multipart(&self, location: &Path) -> OSResult<Box<dyn MultipartUpload>> {
-        let target = self.target.put_multipart(location).await?;
-        Ok(Box::new(IoTrackingMultipartUpload {
-            target,
-            stats: self.stats.clone(),
-        }))
-    }
-
     async fn put_multipart_opts(
         &self,
         location: &Path,
@@ -116,15 +103,6 @@ impl ObjectStore for IoTrackingStore {
         }))
     }
 
-    async fn get(&self, location: &Path) -> OSResult<GetResult> {
-        let result = self.target.get(location).await;
-        if let Ok(result) = &result {
-            let num_bytes = result.range.end - result.range.start;
-            self.record_read(num_bytes);
-        }
-        result
-    }
-
     async fn get_opts(&self, location: &Path, options: GetOptions) -> OSResult<GetResult> {
         let result = self.target.get_opts(location, options).await;
         if let Ok(result) = &result {
@@ -134,14 +112,6 @@ impl ObjectStore for IoTrackingStore {
         result
     }
 
-    async fn get_range(&self, location: &Path, range: std::ops::Range<u64>) -> OSResult<Bytes> {
-        let result = self.target.get_range(location, range).await;
-        if let Ok(result) = &result {
-            self.record_read(result.len() as u64);
-        }
-        result
-    }
-
     async fn get_ranges(
         &self,
         location: &Path,
@@ -154,20 +124,11 @@ impl ObjectStore for IoTrackingStore {
         result
     }
 
-    async fn head(&self, location: &Path) -> OSResult<ObjectMeta> {
+    fn delete_stream(
-        self.record_read(0);
+        &self,
-        self.target.head(location).await
+        locations: BoxStream<'static, OSResult<Path>>,
-    }
+    ) -> BoxStream<'static, OSResult<Path>> {
-
-    async fn delete(&self, location: &Path) -> OSResult<()> {
         self.record_write(0);
-        self.target.delete(location).await
-    }
-
-    fn delete_stream<'a>(
-        &'a self,
-        locations: BoxStream<'a, OSResult<Path>>,
-    ) -> BoxStream<'a, OSResult<Path>> {
         self.target.delete_stream(locations)
     }
 
@@ -190,24 +151,14 @@ impl ObjectStore for IoTrackingStore {
         self.target.list_with_delimiter(prefix).await
     }
 
-    async fn copy(&self, from: &Path, to: &Path) -> OSResult<()> {
+    async fn copy_opts(&self, from: &Path, to: &Path, options: CopyOptions) -> OSResult<()> {
         self.record_write(0);
-        self.target.copy(from, to).await
+        self.target.copy_opts(from, to, options).await
     }
 
-    async fn rename(&self, from: &Path, to: &Path) -> OSResult<()> {
+    async fn rename_opts(&self, from: &Path, to: &Path, options: RenameOptions) -> OSResult<()> {
         self.record_write(0);
-        self.target.rename(from, to).await
+        self.target.rename_opts(from, to, options).await
-    }
-
-    async fn rename_if_not_exists(&self, from: &Path, to: &Path) -> OSResult<()> {
-        self.record_write(0);
-        self.target.rename_if_not_exists(from, to).await
-    }
-
-    async fn copy_if_not_exists(&self, from: &Path, to: &Path) -> OSResult<()> {
-        self.record_write(0);
-        self.target.copy_if_not_exists(from, to).await
     }
 }
 

rust/lancedb/src/remote/client.rs

@@ -16,7 +16,7 @@ use crate::remote::retry::{ResolvedRetryConfig, RetryCounter};
 const REQUEST_ID_HEADER: HeaderName = HeaderName::from_static("x-request-id");
 
 /// Configuration for TLS/mTLS settings.
-#[derive(Clone, Debug, Default)]
+#[derive(Clone, Debug)]
 pub struct TlsConfig {
     /// Path to the client certificate file (PEM format)
     pub cert_file: Option<String>,
@@ -24,10 +24,22 @@ pub struct TlsConfig {
     pub key_file: Option<String>,
     /// Path to the CA certificate file for server verification (PEM format)
     pub ssl_ca_cert: Option<String>,
-    /// Whether to verify the hostname in the server's certificate
+    /// Whether to verify the hostname in the server's certificate.
+    /// Defaults to `true`.
     pub assert_hostname: bool,
 }
 
+impl Default for TlsConfig {
+    fn default() -> Self {
+        Self {
+            cert_file: None,
+            key_file: None,
+            ssl_ca_cert: None,
+            assert_hostname: true,
+        }
+    }
+}
+
 /// Trait for providing custom headers for each request
 #[async_trait::async_trait]
 pub trait HeaderProvider: Send + Sync + std::fmt::Debug {
@@ -926,7 +938,7 @@ mod tests {
         assert!(config.cert_file.is_none());
         assert!(config.key_file.is_none());
         assert!(config.ssl_ca_cert.is_none());
-        assert!(!config.assert_hostname);
+        assert!(config.assert_hostname);
     }
 
     #[test]

rust/lancedb/src/remote/table/insert.rs

@@ -43,7 +43,7 @@ pub struct RemoteInsertExec<S: HttpSend = Sender> {
     client: RestfulLanceDbClient<S>,
     input: Arc<dyn ExecutionPlan>,
     overwrite: bool,
-    properties: PlanProperties,
+    properties: Arc<PlanProperties>,
     add_result: Arc<Mutex<Option<AddResult>>>,
     metrics: ExecutionPlanMetricsSet,
     upload_id: Option<String>,
@@ -118,7 +118,7 @@ impl<S: HttpSend + 'static> RemoteInsertExec<S> {
             client,
             input,
             overwrite,
-            properties,
+            properties: Arc::new(properties),
             add_result: Arc::new(Mutex::new(None)),
             metrics: ExecutionPlanMetricsSet::new(),
             upload_id,
@@ -232,7 +232,7 @@ impl<S: HttpSend + 'static> ExecutionPlan for RemoteInsertExec<S> {
         self
     }
 
-    fn properties(&self) -> &PlanProperties {
+    fn properties(&self) -> &Arc<PlanProperties> {
         &self.properties
     }
 

rust/lancedb/src/table/datafusion.rs

@@ -39,21 +39,26 @@ use lance_index::scalar::FullTextSearchQuery;
 struct MetadataEraserExec {
     input: Arc<dyn ExecutionPlan>,
     schema: Arc<ArrowSchema>,
-    properties: PlanProperties,
+    properties: Arc<PlanProperties>,
 }
 
 impl MetadataEraserExec {
     fn compute_properties_from_input(
         input: &Arc<dyn ExecutionPlan>,
         schema: &Arc<ArrowSchema>,
-    ) -> PlanProperties {
+    ) -> Arc<PlanProperties> {
         let input_properties = input.properties();
         let eq_properties = input_properties
             .eq_properties
             .clone()
             .with_new_schema(schema.clone())
             .unwrap();
-        input_properties.clone().with_eq_properties(eq_properties)
+        Arc::new(
+            input_properties
+                .as_ref()
+                .clone()
+                .with_eq_properties(eq_properties),
+        )
     }
 
     fn new(input: Arc<dyn ExecutionPlan>) -> Self {
@@ -87,7 +92,7 @@ impl ExecutionPlan for MetadataEraserExec {
         self
     }
 
-    fn properties(&self) -> &PlanProperties {
+    fn properties(&self) -> &Arc<PlanProperties> {
         &self.properties
     }
 

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

@@ -81,7 +81,7 @@ pub struct InsertExec {
     dataset: Arc<Dataset>,
     input: Arc<dyn ExecutionPlan>,
     write_params: WriteParams,
-    properties: PlanProperties,
+    properties: Arc<PlanProperties>,
     partial_transactions: Arc<Mutex<Vec<Transaction>>>,
     metrics: ExecutionPlanMetricsSet,
 }
@@ -107,7 +107,7 @@ impl InsertExec {
             dataset,
             input,
             write_params,
-            properties,
+            properties: Arc::new(properties),
             partial_transactions: Arc::new(Mutex::new(Vec::with_capacity(num_partitions))),
             metrics: ExecutionPlanMetricsSet::new(),
         }
@@ -136,7 +136,7 @@ impl ExecutionPlan for InsertExec {
         self
     }
 
-    fn properties(&self) -> &PlanProperties {
+    fn properties(&self) -> &Arc<PlanProperties> {
         &self.properties
     }
 

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

@@ -20,7 +20,7 @@ pub(crate) struct ScannableExec {
     // We don't require Scannable to be Sync, so we wrap it in a Mutex to allow safe concurrent access.
     source: Mutex<Box<dyn Scannable>>,
     num_rows: Option<usize>,
-    properties: PlanProperties,
+    properties: Arc<PlanProperties>,
     tracker: Option<Arc<WriteProgressTracker>>,
 }
 
@@ -49,7 +49,7 @@ impl ScannableExec {
         Self {
             source,
             num_rows,
-            properties,
+            properties: Arc::new(properties),
             tracker,
         }
     }
@@ -70,7 +70,7 @@ impl ExecutionPlan for ScannableExec {
         self
     }
 
-    fn properties(&self) -> &PlanProperties {
+    fn properties(&self) -> &Arc<PlanProperties> {
         &self.properties
     }