fix(python): reject unsupported index accelerators

<!-- lance-gatekeeper-fix:v1 agent=5c80c44c083b3b8ad0da595419d468fc
generation=1 -->

## Root cause

The legacy synchronous Python table called `delete` on a shared, mutable
`lance.Dataset`. Concurrent table operations could hold a PyO3 borrow
while delete requested an exclusive borrow, producing `RuntimeError:
Already borrowed`. The current async-backed binding fixes this by
cloning its thread-safe Rust table handle before awaiting, but that
concurrency contract had no regression coverage.

## Fix

- Document why delete must clone the Rust table handle before entering
its async future.
- Add a barrier-synchronized regression test that deletes distinct rows
through one shared table from eight Python threads.
- Verify every delete commits exactly one row, every commit gets a
distinct version, and no rows remain.

## Validation

- `cargo check --quiet --features remote --tests --examples`
- `cargo fmt --all -- --check`
- `uv run --extra tests --extra dev ruff format --check
python/tests/test_table.py`
- `uv run --extra tests --extra dev ruff check
python/tests/test_table.py`
- `uv run --extra tests --extra dev pytest
python/tests/test_table.py::test_concurrent_deletes_are_thread_safe
python/tests/test_table.py::test_delete
python/tests/test_table.py::test_delete_expr
python/tests/test_table.py::test_delete_expr_async -q` (4 passed)
- Manual stress reproduction: 100 concurrent deletes on one table
completed at versions 2–101 with zero rows remaining.

Fixes #530

Co-authored-by: Gatefixer <313497061+lancedb-gatefixer[bot]@users.noreply.github.com>

python/python/lancedb/_lancedb.pyi

@@ -261,6 +261,7 @@ class Table:
     def name(self) -> str: ...
     def __repr__(self) -> str: ...
     def is_open(self) -> bool: ...
+    def _is_native(self) -> bool: ...
     def close(self) -> None: ...
     async def schema(self) -> pa.Schema: ...
     async def add(

python/python/lancedb/db.py

@@ -707,6 +707,9 @@ class LanceDBConnection(DBConnection):
         self._namespace_client_properties = namespace_client_properties
         if _inner is not None:
             self._conn = _inner
+            # Native-derived wrappers resolve this in their async reconstruction
+            # path so construction never synchronously re-enters LOOP.
+            self._read_consistency_interval = read_consistency_interval
             self._cached_namespace_client = None
             return
 
@@ -756,11 +759,14 @@ class LanceDBConnection(DBConnection):
         # storage_options.  Also, this class really shouldn't be holding any state
         # beyond _conn.
         self._conn = AsyncConnection(LOOP.run(do_connect()))
+        # Keep property access synchronous so debugger introspection cannot wait on
+        # the background loop while that thread is suspended at a breakpoint.
+        self._read_consistency_interval = read_consistency_interval
         self._cached_namespace_client: Optional[LanceNamespace] = None
 
     @property
     def read_consistency_interval(self) -> Optional[timedelta]:
-        return LOOP.run(self._conn.get_read_consistency_interval())
+        return self._read_consistency_interval
 
     @property
     def session(self) -> Optional[Session]:
@@ -771,8 +777,16 @@ class LanceDBConnection(DBConnection):
         return self._conn.uri
 
     @classmethod
-    def from_inner(cls, inner: LanceDbConnection):
-        return cls(None, _inner=inner)
+    def from_inner(
+        cls,
+        inner: LanceDbConnection,
+        read_consistency_interval: Optional[timedelta],
+    ):
+        return cls(
+            None,
+            read_consistency_interval=read_consistency_interval,
+            _inner=inner,
+        )
 
     def __repr__(self) -> str:
         return f"{self.__class__.__name__}(uri={self._conn.uri!r})"

python/python/lancedb/index.py

@@ -314,8 +314,7 @@ class HnswPq:
     m: int = 20
     ef_construction: int = 300
     target_partition_size: Optional[int] = None
-    # Name of the accelerator (e.g. "cuda") to use for IVF training. When set,
-    # create_index() dispatches to pylance to build the index on the accelerator.
+    # Reserved for future accelerator support. create_index() currently raises if set.
     accelerator: Optional[str] = None
 
 
@@ -422,8 +421,7 @@ class HnswSq:
     m: int = 20
     ef_construction: int = 300
     target_partition_size: Optional[int] = None
-    # Name of the accelerator (e.g. "cuda") to use for IVF training. When set,
-    # create_index() dispatches to pylance to build the index on the accelerator.
+    # Reserved for future accelerator support. create_index() currently raises if set.
     accelerator: Optional[str] = None
 
 
@@ -618,8 +616,7 @@ class IvfFlat:
     max_iterations: int = 50
     sample_rate: int = 256
     target_partition_size: Optional[int] = None
-    # Name of the accelerator (e.g. "cuda") to use for IVF training. When set,
-    # create_index() dispatches to pylance to build the index on the accelerator.
+    # Reserved for future accelerator support. create_index() currently raises if set.
     accelerator: Optional[str] = None
 
 
@@ -651,8 +648,7 @@ class IvfSq:
     max_iterations: int = 50
     sample_rate: int = 256
     target_partition_size: Optional[int] = None
-    # Name of the accelerator (e.g. "cuda") to use for IVF training. When set,
-    # create_index() dispatches to pylance to build the index on the accelerator.
+    # Reserved for future accelerator support. create_index() currently raises if set.
     accelerator: Optional[str] = None
 
 
@@ -784,7 +780,7 @@ class IvfPq:
     max_iterations: int = 50
     sample_rate: int = 256
     target_partition_size: Optional[int] = None
-    # Name of the accelerator (e.g. "cuda") to use for IVF training. When set,
+    # Name of the accelerator ("cuda" or "mps") to use for IVF training. When set,
     # create_index() dispatches to pylance to build the index on the accelerator.
     accelerator: Optional[str] = None
 
@@ -840,8 +836,7 @@ class IvfRq:
     max_iterations: int = 50
     sample_rate: int = 256
     target_partition_size: Optional[int] = None
-    # Name of the accelerator (e.g. "cuda") to use for IVF training. When set,
-    # create_index() dispatches to pylance to build the index on the accelerator.
+    # Reserved for future accelerator support. create_index() currently raises if set.
     accelerator: Optional[str] = None
 
 

python/python/lancedb/permutation.py

@@ -226,7 +226,7 @@ class PermutationBuilder:
 
         async def do_execute():
             inner_tbl = await self._async.execute()
-            return LanceTable.from_inner(inner_tbl)
+            return await LanceTable.from_inner(inner_tbl)
 
         return LOOP.run(do_execute())
 

python/python/lancedb/remote/table.py

@@ -68,6 +68,14 @@ from ..table import AsyncTable, BlobMode, Branches, IndexStatistics, Query, Tabl
 from ..types import BaseTokenizerType
 
 
+def _reject_index_accelerator(
+    config: Optional[IndexConfigType] = None,
+    accelerator: Optional[str] = None,
+) -> None:
+    if accelerator is not None or getattr(config, "accelerator", None) is not None:
+        raise ValueError("Index accelerators are not supported on LanceDB Cloud.")
+
+
 class RemoteTable(Table):
     def __init__(
         self,
@@ -457,6 +465,8 @@ class RemoteTable(Table):
         ...     "l2", vector_column_name="vector"
         ... )
         """
+        _reject_index_accelerator(config, accelerator)
+
         # Detect whether this is a legacy API call
         is_legacy = self._is_legacy_create_index_call(
             metric,
@@ -484,12 +494,6 @@ class RemoteTable(Table):
 
             column = vector_column_name
 
-            if accelerator is not None:
-                logging.warning(
-                    "GPU accelerator is not yet supported on LanceDB cloud."
-                    "If you have 100M+ vectors to index,"
-                    "please contact us at contact@lancedb.com"
-                )
             if replace is not None:
                 logging.warning(
                     "replace is not supported on LanceDB cloud."
@@ -557,6 +561,8 @@ class RemoteTable(Table):
         The job may already be complete when returned; callers must not assume
         the index exists until :meth:`Job.wait` returns.
         """
+        _reject_index_accelerator(config)
+
         return Job(
             LOOP.run(
                 self._table.create_index_async(

python/python/lancedb/table.py

@@ -214,6 +214,45 @@ IndexConfigType = Union[
 # Known distance metrics for legacy API detection
 KNOWN_METRICS = {"l2", "cosine", "dot", "hamming"}
 
+_PYLANCE_ACCELERATED_INDEX_TYPE = "IVF_PQ"
+
+
+def _pylance_accelerated_index_options(
+    config: IndexConfigType,
+    *,
+    accelerator: Optional[str] = None,
+    index_type: Optional[str] = None,
+) -> Optional[Dict[str, Any]]:
+    """Translate an accelerated vector config into PyLance index options."""
+    if accelerator is None:
+        accelerator = getattr(config, "accelerator", None)
+    if accelerator is None:
+        return None
+
+    if index_type is None:
+        index_type = (
+            _PYLANCE_ACCELERATED_INDEX_TYPE
+            if isinstance(config, IvfPq)
+            else type(config).__name__
+        )
+    if index_type.upper() != _PYLANCE_ACCELERATED_INDEX_TYPE:
+        raise ValueError(
+            f"Index type {index_type} does not support an accelerator; "
+            f"only {_PYLANCE_ACCELERATED_INDEX_TYPE} supports acceleration"
+        )
+
+    return {
+        "index_type": index_type,
+        "metric": getattr(config, "distance_type", "l2"),
+        "num_partitions": getattr(config, "num_partitions", None),
+        "num_sub_vectors": getattr(config, "num_sub_vectors", None),
+        "accelerator": accelerator,
+        "num_bits": getattr(config, "num_bits", 8),
+        "m": getattr(config, "m", 20),
+        "ef_construction": getattr(config, "ef_construction", 300),
+        "target_partition_size": getattr(config, "target_partition_size", None),
+    }
+
 
 def _into_pyarrow_reader(
     data, schema: Optional[pa.Schema] = None
@@ -2182,11 +2221,15 @@ class LanceTable(Table):
         return self.name
 
     @classmethod
-    def from_inner(cls, tbl: LanceDBTable):
-        from .db import LanceDBConnection
+    async def from_inner(cls, tbl: LanceDBTable):
+        from .db import AsyncConnection, LanceDBConnection
 
         async_tbl = AsyncTable(tbl)
-        conn = LanceDBConnection.from_inner(tbl.database())
+        inner_conn = tbl.database()
+        read_consistency_interval = await AsyncConnection(
+            inner_conn
+        ).get_read_consistency_interval()
+        conn = LanceDBConnection.from_inner(inner_conn, read_consistency_interval)
         return cls(
             conn,
             async_tbl.name,
@@ -2733,20 +2776,17 @@ class LanceTable(Table):
             )
 
             # Handle accelerator through pylance
-            if accelerator is not None:
+            accelerated_options = _pylance_accelerated_index_options(
+                config, accelerator=accelerator, index_type=index_type
+            )
+            if accelerated_options is not None:
                 self.to_lance().create_index(
                     column=column,
-                    index_type=index_type,
-                    metric=metric,
-                    num_partitions=num_partitions,
-                    num_sub_vectors=num_sub_vectors,
                     replace=replace,
-                    accelerator=accelerator,
                     index_cache_size=index_cache_size,
-                    num_bits=num_bits,
-                    m=m,
-                    ef_construction=ef_construction,
-                    target_partition_size=target_partition_size,
+                    name=name,
+                    train=train,
+                    **accelerated_options,
                 )
                 self.checkout_latest()
                 return
@@ -2754,39 +2794,21 @@ class LanceTable(Table):
             # New API: metric is the column name
             column = metric
 
-            # Check if config has accelerator set and dispatch to pylance
-            if config is not None and hasattr(config, "accelerator"):
-                acc = getattr(config, "accelerator", None)
-                if acc is not None:
-                    # Dispatch to pylance for GPU acceleration
-                    index_type_map = {
-                        "IvfFlat": "IVF_FLAT",
-                        "IvfSq": "IVF_SQ",
-                        "IvfPq": "IVF_PQ",
-                        "IvfRq": "IVF_RQ",
-                        "HnswPq": "IVF_HNSW_PQ",
-                        "HnswSq": "IVF_HNSW_SQ",
-                    }
-                    cfg_type = type(config).__name__
-                    lance_index_type = index_type_map.get(cfg_type, "IVF_PQ")
-
-                    self.to_lance().create_index(
-                        column=column,
-                        index_type=lance_index_type,
-                        metric=getattr(config, "distance_type", "l2"),
-                        num_partitions=getattr(config, "num_partitions", None),
-                        num_sub_vectors=getattr(config, "num_sub_vectors", None),
-                        replace=replace,
-                        accelerator=acc,
-                        num_bits=getattr(config, "num_bits", 8),
-                        m=getattr(config, "m", 20),
-                        ef_construction=getattr(config, "ef_construction", 300),
-                        target_partition_size=getattr(
-                            config, "target_partition_size", None
-                        ),
-                    )
-                    self.checkout_latest()
-                    return
+            accelerated_options = (
+                _pylance_accelerated_index_options(config)
+                if config is not None
+                else None
+            )
+            if accelerated_options is not None:
+                self.to_lance().create_index(
+                    column=column,
+                    replace=replace,
+                    name=name,
+                    train=train,
+                    **accelerated_options,
+                )
+                self.checkout_latest()
+                return
 
         return LOOP.run(
             self._table.create_index(
@@ -2814,6 +2836,11 @@ class LanceTable(Table):
         The job may already be complete when returned; callers must not assume
         the index exists until :meth:`Job.wait` returns.
         """
+        if _pylance_accelerated_index_options(config) is not None:
+            raise ValueError(
+                "Accelerated index creation does not support create_index_async; "
+                "use create_index instead."
+            )
         return Job(
             LOOP.run(
                 self._table.create_index_async(
@@ -4826,6 +4853,7 @@ class AsyncTable:
         config: Optional[
             Union[
                 IvfFlat,
+                IvfSq,
                 IvfPq,
                 IvfRq,
                 HnswPq,
@@ -4896,6 +4924,23 @@ class AsyncTable:
                     " BTree, Bitmap, LabelList, Fm, or FTS, but got "
                     + str(type(config))
                 )
+        accelerated_options = (
+            _pylance_accelerated_index_options(config) if config is not None else None
+        )
+        if accelerated_options is not None:
+            if not self._inner._is_native():
+                raise ValueError("GPU accelerator is not supported on LanceDB Cloud.")
+            dataset = await self.to_lance()
+            await asyncio.to_thread(
+                dataset.create_index,
+                column=column,
+                replace=True if replace is None else replace,
+                name=name,
+                train=train,
+                **accelerated_options,
+            )
+            await self.checkout_latest()
+            return
         try:
             await self._inner.create_index(
                 column,
@@ -4922,6 +4967,7 @@ class AsyncTable:
         config: Optional[
             Union[
                 IvfFlat,
+                IvfSq,
                 IvfPq,
                 IvfRq,
                 HnswPq,
@@ -4944,6 +4990,11 @@ class AsyncTable:
         be complete when returned; callers must not assume the index exists
         until :meth:`AsyncJob.wait` resolves.
         """
+        if config is not None and _pylance_accelerated_index_options(config):
+            raise ValueError(
+                "Accelerated index creation does not support create_index_async; "
+                "use create_index instead."
+            )
         job = await self._inner.create_index_async(
             column,
             index=config,

python/python/tests/test_db.py

@@ -77,6 +77,23 @@ def test_sync_repr_does_not_use_background_loop(tmp_path, monkeypatch):
     assert repr(table) == f"LanceTable(name='test', _conn={db!r})"
 
 
+def test_read_consistency_interval_does_not_use_background_loop(tmp_path, monkeypatch):
+    from lancedb.background_loop import LOOP
+    from lancedb.db import LanceDBConnection
+
+    consistency_interval = timedelta(seconds=5)
+    db = lancedb.connect(tmp_path, read_consistency_interval=consistency_interval)
+    db_from_inner = LanceDBConnection.from_inner(db._inner, consistency_interval)
+
+    def fail_run(*args, **kwargs):
+        raise AssertionError("properties should not use the Python background loop")
+
+    monkeypatch.setattr(LOOP, "run", fail_run)
+
+    assert db.read_consistency_interval == consistency_interval
+    assert db_from_inner.read_consistency_interval == consistency_interval
+
+
 def test_ingest_pd(tmp_path):
     db = lancedb.connect(tmp_path)
 

python/python/tests/test_permutation.py

@@ -6,6 +6,7 @@ import math
 import pytest
 
 from lancedb import DBConnection, Table, connect
+from lancedb.background_loop import LOOP
 from lancedb.permutation import Permutation, Permutations, permutation_builder
 
 
@@ -31,6 +32,25 @@ def test_split_random_ratios(mem_db):
     assert 65 <= split_1_count <= 75  # ~70% ± tolerance
 
 
+def test_execute_does_not_reenter_background_loop(tmp_path, monkeypatch):
+    import threading
+
+    db = connect(tmp_path)
+    tbl = db.create_table("test_table", pa.table({"x": range(10)}))
+    original_run = LOOP.run
+
+    def fail_on_reentry(future):
+        assert threading.current_thread() is not LOOP.thread
+        return original_run(future)
+
+    monkeypatch.setattr(LOOP, "run", fail_on_reentry)
+
+    permutation_tbl = permutation_builder(tbl).execute()
+
+    assert permutation_tbl.count_rows() == 10
+    assert permutation_tbl._conn.read_consistency_interval is None
+
+
 def test_split_random_counts(mem_db):
     """Test random splitting with absolute counts."""
     tbl = mem_db.create_table(

python/python/tests/test_remote_db.py

@@ -875,6 +875,25 @@ def test_remote_create_index_async_returns_job():
         job.cancel()
 
 
+def test_remote_create_index_rejects_accelerator():
+    from lancedb.index import IvfPq
+    from lancedb.remote.table import RemoteTable
+
+    inner = MagicMock()
+    inner.name = "test"
+    table = RemoteTable(inner, "dev")
+
+    with pytest.raises(ValueError, match="not supported on LanceDB Cloud"):
+        table.create_index(accelerator="mps")
+    with pytest.raises(ValueError, match="not supported on LanceDB Cloud"):
+        table.create_index("vector", config=IvfPq(accelerator="mps"))
+    with pytest.raises(ValueError, match="not supported on LanceDB Cloud"):
+        table.create_index_async("vector", config=IvfPq(accelerator="mps"))
+
+    inner.create_index.assert_not_called()
+    inner.create_index_async.assert_not_called()
+
+
 def test_remote_job_wait_raises_on_failure():
     from lancedb.exceptions import JobFailedError
     from lancedb.index import BTree

python/python/tests/test_table.py

@@ -6,14 +6,25 @@ import os
 import sys
 import threading
 import warnings
+from concurrent.futures import ThreadPoolExecutor
 from datetime import date, datetime, timedelta
 from time import sleep
 from typing import List
-from unittest.mock import patch
+from unittest.mock import AsyncMock, MagicMock, patch
 
 import lancedb
 from lancedb.dependencies import _PANDAS_AVAILABLE
-from lancedb.index import BTree, FTS, HnswFlat, HnswPq, HnswSq, IvfPq
+from lancedb.index import (
+    BTree,
+    FTS,
+    HnswFlat,
+    HnswPq,
+    HnswSq,
+    IvfFlat,
+    IvfPq,
+    IvfRq,
+    IvfSq,
+)
 import numpy as np
 import polars as pl
 import pyarrow as pa
@@ -24,7 +35,7 @@ from lancedb.db import AsyncConnection, DBConnection
 from lancedb.embeddings import EmbeddingFunctionConfig, EmbeddingFunctionRegistry
 from lancedb.expr import col, lit
 from lancedb.pydantic import LanceModel, Vector
-from lancedb.table import LanceTable
+from lancedb.table import AsyncTable, LanceTable
 from pydantic import BaseModel
 
 
@@ -1411,6 +1422,174 @@ def test_create_index_async_returns_done_job(mem_db: DBConnection):
     job.cancel()
 
 
+def test_create_index_dispatches_mps_to_pylance(mem_db: DBConnection):
+    table = mem_db.create_table(
+        "mps_sync",
+        data=[
+            {"vector": [3.1, 4.1]},
+            {"vector": [5.9, 26.5]},
+        ],
+    )
+    dataset = MagicMock()
+
+    with (
+        patch.object(table, "to_lance", return_value=dataset),
+        patch.object(table, "checkout_latest") as checkout_latest,
+    ):
+        with pytest.warns(DeprecationWarning, match="create_index"):
+            table.create_index(
+                metric="cosine",
+                num_partitions=4,
+                num_sub_vectors=2,
+                accelerator="mps",
+                replace=False,
+                name="vector_mps",
+            )
+
+    dataset.create_index.assert_called_once_with(
+        column="vector",
+        replace=False,
+        index_cache_size=None,
+        name="vector_mps",
+        train=True,
+        index_type="IVF_PQ",
+        metric="cosine",
+        num_partitions=4,
+        num_sub_vectors=2,
+        accelerator="mps",
+        num_bits=8,
+        m=20,
+        ef_construction=300,
+        target_partition_size=None,
+    )
+    checkout_latest.assert_called_once_with()
+
+
+@pytest.mark.parametrize(
+    "config",
+    [
+        IvfFlat(accelerator="mps"),
+        IvfSq(accelerator="mps"),
+        IvfRq(accelerator="mps"),
+        HnswPq(accelerator="mps"),
+        HnswSq(accelerator="mps"),
+    ],
+)
+def test_create_index_rejects_unsupported_accelerated_format(
+    mem_db: DBConnection, config
+):
+    table = mem_db.create_table(
+        "unsupported_accelerator",
+        data=[{"vector": [3.1, 4.1]}, {"vector": [5.9, 26.5]}],
+    )
+
+    with (
+        patch.object(table, "to_lance") as to_lance,
+        pytest.raises(ValueError, match="only IVF_PQ supports acceleration"),
+    ):
+        table.create_index("vector", config=config)
+
+    to_lance.assert_not_called()
+
+
+def test_legacy_create_index_rejects_unsupported_accelerated_format(
+    mem_db: DBConnection,
+):
+    table = mem_db.create_table(
+        "unsupported_legacy_accelerator",
+        data=[{"vector": [3.1, 4.1]}, {"vector": [5.9, 26.5]}],
+    )
+
+    with (
+        pytest.warns(DeprecationWarning, match="create_index"),
+        patch.object(table, "to_lance") as to_lance,
+        pytest.raises(ValueError, match="only IVF_PQ supports acceleration"),
+    ):
+        table.create_index(index_type="IVF_FLAT", accelerator="mps")
+
+    to_lance.assert_not_called()
+
+
+@pytest.mark.asyncio
+async def test_async_create_index_dispatches_mps_to_pylance():
+    inner = MagicMock()
+    inner._is_native.return_value = True
+    inner.checkout_latest = AsyncMock()
+    table = AsyncTable(inner)
+    dataset = MagicMock()
+
+    with patch.object(table, "to_lance", AsyncMock(return_value=dataset)):
+        await table.create_index(
+            "vector",
+            config=IvfPq(
+                distance_type="cosine",
+                num_partitions=4,
+                num_sub_vectors=2,
+                accelerator="mps",
+            ),
+            name="vector_mps",
+        )
+
+    dataset.create_index.assert_called_once_with(
+        column="vector",
+        replace=True,
+        name="vector_mps",
+        train=True,
+        index_type="IVF_PQ",
+        metric="cosine",
+        num_partitions=4,
+        num_sub_vectors=2,
+        accelerator="mps",
+        num_bits=8,
+        m=20,
+        ef_construction=300,
+        target_partition_size=None,
+    )
+    inner.create_index.assert_not_called()
+    inner.checkout_latest.assert_awaited_once_with()
+
+
+@pytest.mark.asyncio
+async def test_async_create_index_rejects_unsupported_accelerated_format():
+    inner = MagicMock()
+    inner._is_native.return_value = True
+    table = AsyncTable(inner)
+
+    with (
+        patch.object(table, "to_lance", AsyncMock()) as to_lance,
+        pytest.raises(ValueError, match="only IVF_PQ supports acceleration"),
+    ):
+        await table.create_index("vector", config=IvfFlat(accelerator="mps"))
+
+    to_lance.assert_not_awaited()
+    inner.create_index.assert_not_called()
+
+
+@pytest.mark.asyncio
+async def test_async_background_index_rejects_accelerator():
+    inner = MagicMock()
+    inner.create_index_async = AsyncMock()
+    table = AsyncTable(inner)
+
+    with pytest.raises(ValueError, match="Accelerated index creation does not support"):
+        await table.create_index_async("vector", config=IvfPq(accelerator="mps"))
+
+    inner.create_index_async.assert_not_awaited()
+
+
+def test_background_index_rejects_accelerator(mem_db: DBConnection):
+    table = mem_db.create_table(
+        "mps_background",
+        data=[
+            {"vector": [3.1, 4.1]},
+            {"vector": [5.9, 26.5]},
+        ],
+    )
+
+    with pytest.raises(ValueError, match="Accelerated index creation does not support"):
+        table.create_index_async("vector", config=IvfPq(accelerator="mps"))
+
+
 @patch("lancedb.table.AsyncTable.create_index")
 def test_create_index_method(mock_create_index, mem_db: DBConnection):
     table = mem_db.create_table(
@@ -2124,6 +2303,27 @@ def test_delete(mem_db: DBConnection):
     assert table.to_arrow()["id"].to_pylist() == [1]
 
 
+def test_concurrent_deletes_are_thread_safe(mem_db: DBConnection):
+    num_workers = 8
+    table = mem_db.create_table(
+        "my_table", data=[{"id": row_id} for row_id in range(num_workers)]
+    )
+    barrier = threading.Barrier(num_workers)
+
+    def delete(row_id: int):
+        barrier.wait()
+        return table.delete(f"id = {row_id}")
+
+    with ThreadPoolExecutor(max_workers=num_workers) as pool:
+        results = list(pool.map(delete, range(num_workers)))
+
+    assert all(result.num_deleted_rows == 1 for result in results)
+    assert sorted(result.version for result in results) == list(
+        range(2, num_workers + 2)
+    )
+    assert table.count_rows() == 0
+
+
 def test_delete_expr(mem_db: DBConnection):
     table = mem_db.create_table(
         "my_table",

python/src/table.rs

@@ -622,6 +622,10 @@ impl Table {
         self.inner.is_some()
     }
 
+    pub fn _is_native(&self) -> PyResult<bool> {
+        Ok(self.inner_ref()?.as_native().is_some())
+    }
+
     /// Closes the table, releasing any resources associated with it.
     pub fn close(&mut self) {
         self.inner.take();
@@ -745,6 +749,9 @@ impl Table {
 
     #[allow(private_interfaces)]
     pub fn delete(self_: PyRef<'_, Self>, condition: PredicateArg) -> PyResult<Bound<'_, PyAny>> {
+        // Do not hold the Python borrow across the await.  The cloned Rust table
+        // handle is thread-safe and allows deletes on the same Python table to
+        // run concurrently without PyO3 reporting "Already borrowed".
         let inner = self_.inner_ref()?.clone();
         future_into_py(self_.py(), async move {
             let result = match &condition {