Fix analysis sandbox "Already borrowed" crash on VFS reads

CHANGELOG.md

@@ -9,6 +9,7 @@
 
 - Opening a database no longer writes to it: reads no longer rewrite the stored embedding settings or change the stored version, and the version is never downgraded. Embedding provider/name drift (matching `vector_dim`) warns on read-only opens and raises `ConfigMismatchError` on writable opens; reconcile with `rebuild --set-embedder`.
 - Read CLI verbs (`list`, `get`, `search`, `visualize`, `ask`, `analyze`, `inspect`, `chat`, `info`, `history`) open the database read-only.
+- Analysis-skill `execute_code` no longer crashes with LanceDB `Already borrowed` when sandboxed code reads the document VFS (`content.txt`, `items.jsonl`, `toc.json`); reads run on one background event loop with a single read-only connection for the sandbox's lifetime.
 
 ## [0.54.0] - 2026-06-04
 

haiku_rag_slim/haiku/rag/sandbox/sandbox.py

@@ -1,9 +1,9 @@
 import asyncio
-import atexit
-import concurrent.futures
 import json
 import os
-from collections.abc import Callable
+import threading
+import weakref
+from collections.abc import Callable, Coroutine
 from dataclasses import dataclass
 from pathlib import Path
 from typing import TYPE_CHECKING, Any, Literal
@@ -19,6 +19,8 @@ from haiku.rag.store.models.document_item import PICTURE_REF_PREFIX, DocumentIte
 if TYPE_CHECKING:
     from pathlib import PurePosixPath
 
+    from haiku.rag.client import HaikuRAG
+
 
 @dataclass
 class SandboxResult:
@@ -29,13 +31,13 @@ class SandboxResult:
     success: bool
 
 
-_executor = concurrent.futures.ThreadPoolExecutor(max_workers=1)
+def _stop_loop(loop: asyncio.AbstractEventLoop, thread: threading.Thread) -> None:
-atexit.register(_executor.shutdown, wait=False)
+    """Stop a background loop and join its thread. Safe if already stopped."""
-
+    if loop.is_closed():
-
+        return
-def _run_async(coro: Any) -> Any:
+    loop.call_soon_threadsafe(loop.stop)
-    """Run an async coroutine from a sync context (CallbackFile read)."""
+    thread.join()
-    return _executor.submit(asyncio.run, coro).result()
+    loop.close()
 
 
 def _build_toc(
@@ -127,6 +129,13 @@ class Sandbox:
         sandbox = Sandbox(db_path, config, context)
         result = await sandbox.execute("x = await search('query')")
         result = await sandbox.execute("print(x[0]['content'])")  # x persists
+
+    The virtual filesystem reads run on a dedicated background event loop with a
+    single read-only connection held for the sandbox's lifetime. Monty's file
+    callbacks are synchronous and cannot ``await``, so they bridge to that loop
+    via ``run_coroutine_threadsafe``. Call ``close()`` to release the loop and
+    connection; if it is not called they are released when the instance is
+    garbage collected.
     """
 
     _db_path: Path
@@ -139,6 +148,10 @@ class Sandbox:
     _toc_json_cache: dict[str, str]
     _repl: MontyRepl | None
     _vfs: OSAccess | None
+    _loop: asyncio.AbstractEventLoop | None
+    _loop_thread: threading.Thread | None
+    _vfs_rag: "HaikuRAG | None"
+    _finalizer: weakref.finalize | None
 
     def __init__(
         self,
@@ -156,6 +169,60 @@ class Sandbox:
         self._toc_json_cache = {}
         self._repl = None
         self._vfs = None
+        self._loop = None
+        self._loop_thread = None
+        self._vfs_rag = None
+        self._finalizer = None
+
+    def _ensure_vfs_loop(self) -> None:
+        """Start the background loop and open the read-only connection once."""
+        if self._loop is not None:
+            return
+        from haiku.rag.client import HaikuRAG
+
+        loop = asyncio.new_event_loop()
+        thread = threading.Thread(
+            target=loop.run_forever, daemon=True, name="sandbox-vfs"
+        )
+        thread.start()
+        self._loop = loop
+        self._loop_thread = thread
+
+        async def _open() -> "HaikuRAG":
+            rag = HaikuRAG(self._db_path, config=self._config, read_only=True)
+            await rag.__aenter__()
+            return rag
+
+        self._vfs_rag = self._run_on_loop(_open())
+        # GC backstop: capture only the loop + thread, never ``self`` (which
+        # would pin the instance and prevent collection).
+        self._finalizer = weakref.finalize(self, _stop_loop, loop, thread)
+
+    def _run_on_loop(self, coro: Coroutine[Any, Any, Any]) -> Any:
+        """Run a coroutine on the background loop from a synchronous caller."""
+        self._ensure_vfs_loop()
+        assert self._loop is not None
+        return asyncio.run_coroutine_threadsafe(coro, self._loop).result()
+
+    def close(self) -> None:
+        """Release the background loop and read-only connection.
+
+        Idempotent and safe to call when no VFS read ever started the loop.
+        """
+        loop, thread, rag = self._loop, self._loop_thread, self._vfs_rag
+        if loop is None or thread is None:
+            return
+        self._loop = None
+        self._loop_thread = None
+        self._vfs_rag = None
+        if self._finalizer is not None:
+            self._finalizer.detach()
+            self._finalizer = None
+        if rag is not None and not loop.is_closed():
+            asyncio.run_coroutine_threadsafe(
+                rag.__aexit__(None, None, None), loop
+            ).result()
+        _stop_loop(loop, thread)
 
     def _build_external_functions(self) -> dict[str, Any]:
         """Build async external functions for the Monty interpreter."""
@@ -249,12 +316,12 @@ class Sandbox:
                 return cached
 
             async def _fetch() -> list[DocumentItem]:
-                from haiku.rag.client import HaikuRAG
+                assert sandbox._vfs_rag is not None
+                return await sandbox._vfs_rag.document_item_repository.get_all_items(
+                    did
+                )
 
-                async with HaikuRAG(db_path, config=config, read_only=True) as rag:
+            items = sandbox._run_on_loop(_fetch())
-                    return await rag.document_item_repository.get_all_items(did)
-
-            items = _run_async(_fetch())
             sandbox._doc_items[did] = items
             return items
 
@@ -265,17 +332,13 @@ class Sandbox:
                 return cached
 
             async def _fetch() -> dict[str, list[str]]:
-                from haiku.rag.client import HaikuRAG
+                assert sandbox._vfs_rag is not None
-
+                index = await sandbox._vfs_rag.chunk_repository.get_chunk_ids_by_self_ref_grouped(
-                async with HaikuRAG(db_path, config=config, read_only=True) as rag:
+                    [did]
-                    index = (
+                )
-                        await rag.chunk_repository.get_chunk_ids_by_self_ref_grouped(
-                            [did]
-                        )
-                    )
                 return index.get(did, {})
 
-            chunk_index = _run_async(_fetch())
+            chunk_index = sandbox._run_on_loop(_fetch())
             sandbox._doc_chunk_index[did] = chunk_index
             return chunk_index
 
@@ -351,15 +414,13 @@ class Sandbox:
             ) -> Callable[["PurePosixPath"], str]:
                 def read_content(_path: "PurePosixPath") -> str:
                     async def _fetch() -> str:
-                        from haiku.rag.client import HaikuRAG
+                        assert sandbox._vfs_rag is not None
+                        content = (
+                            await sandbox._vfs_rag.document_repository.get_content(did)
+                        )
+                        return content or ""
 
-                        async with HaikuRAG(
+                    return sandbox._run_on_loop(_fetch())
-                            db_path, config=config, read_only=True
-                        ) as rag:
-                            content = await rag.document_repository.get_content(did)
-                            return content or ""
-
-                    return _run_async(_fetch())
 
                 return read_content
 

haiku_rag_slim/haiku/rag/skills/_deps.py

@@ -67,12 +67,16 @@ def make_analysis_lifespan(db_path: Path, config: AppConfig):
         async with HaikuRAG(db_path, config=config, read_only=True) as rag:
             deps.rag = rag
             deps.search_count = 0
-            deps.sandbox = Sandbox(
+            sandbox = Sandbox(
                 db_path=db_path,
                 config=config,
                 context=AnalysisContext(filter=doc_filter),
             )
+            deps.sandbox = sandbox
             _reset_invocation_state(deps.state)
-            yield
+            try:
+                yield
+            finally:
+                sandbox.close()
 
     return lifespan

tests/sandbox/test_sandbox.py

@@ -412,3 +412,136 @@ class TestSandboxVFS:
             assert "1" in result.stdout
             assert "Public Doc" in result.stdout
             assert "Private Doc" not in result.stdout
+
+
+class TestSandboxHeldConnection:
+    """VFS reads must work while another connection to the same DB stays open.
+
+    Mirrors the analysis-skill lifespan, which keeps a read-only connection open
+    for the whole turn while sandboxed code reads the document VFS.
+    """
+
+    @pytest.mark.asyncio
+    @pytest.mark.vcr()
+    async def test_vfs_reads_while_connection_held(self, temp_db_path):
+        """All three VFS readers work while another connection to the DB is open."""
+        config = AppConfig()
+        async with HaikuRAG(temp_db_path, create=True) as client:
+            doc = await client.create_document(
+                content="Foxes and dogs roam the quiet hills.",
+                uri="test://doc",
+                title="Doc",
+            )
+        assert doc.id
+
+        async with HaikuRAG(temp_db_path, config=config, read_only=True):
+            context = AnalysisContext()
+            sb = Sandbox(db_path=temp_db_path, config=config, context=context)
+            try:
+                result = await sb.execute(
+                    "from pathlib import Path\n"
+                    f"d = Path('/documents/{doc.id}')\n"
+                    "print('foxes' in (d / 'content.txt').read_text().lower())\n"
+                    "print(len((d / 'items.jsonl').read_text().strip().split('\\n')))\n"
+                    "print('tree' in (d / 'toc.json').read_text())"
+                )
+                assert result.success, result.stderr
+                lines = result.stdout.strip().split("\n")
+                assert lines[0] == "True"
+                assert int(lines[1]) > 0
+                assert lines[2] == "True"
+            finally:
+                sb.close()
+
+    @pytest.mark.asyncio
+    @pytest.mark.vcr()
+    async def test_vfs_loop_and_connection_are_reused(self, temp_db_path):
+        """The background loop + read-only connection are created once and reused."""
+        config = AppConfig()
+        async with HaikuRAG(temp_db_path, create=True) as client:
+            doc = await client.create_document(
+                content="Foxes and dogs.",
+                uri="test://doc",
+                title="Doc",
+            )
+        assert doc.id
+
+        context = AnalysisContext()
+        sb = Sandbox(db_path=temp_db_path, config=config, context=context)
+        try:
+            assert sb._loop is None and sb._vfs_rag is None
+            read = (
+                "from pathlib import Path\n"
+                f"print(Path('/documents/{doc.id}/content.txt').read_text())"
+            )
+            assert (await sb.execute(read)).success
+            loop, rag = sb._loop, sb._vfs_rag
+            assert loop is not None and rag is not None
+            assert (await sb.execute(read)).success
+            assert sb._loop is loop
+            assert sb._vfs_rag is rag
+        finally:
+            sb.close()
+
+    @pytest.mark.asyncio
+    @pytest.mark.vcr()
+    async def test_variables_persist_across_executes_after_vfs_read(self, temp_db_path):
+        """REPL state persists across execute() calls, including after a VFS read."""
+        config = AppConfig()
+        async with HaikuRAG(temp_db_path, create=True) as client:
+            doc = await client.create_document(
+                content="Foxes and dogs.",
+                uri="test://doc",
+                title="Doc",
+            )
+        assert doc.id
+
+        context = AnalysisContext()
+        sb = Sandbox(db_path=temp_db_path, config=config, context=context)
+        try:
+            first = await sb.execute(
+                "from pathlib import Path\n"
+                f"x = len(Path('/documents/{doc.id}/content.txt').read_text())"
+            )
+            assert first.success, first.stderr
+            second = await sb.execute("print(x)")
+            assert second.success, second.stderr
+            assert int(second.stdout.strip()) > 0
+        finally:
+            sb.close()
+
+    @pytest.mark.asyncio
+    async def test_close_is_safe_without_vfs_read(self, temp_db_path):
+        """close() is a no-op (and safe to call twice) when no VFS read happened."""
+        async with HaikuRAG(temp_db_path, create=True):
+            config = AppConfig()
+            sb = Sandbox(db_path=temp_db_path, config=config, context=AnalysisContext())
+            assert sb._loop is None
+            sb.close()
+            sb.close()
+
+    @pytest.mark.asyncio
+    @pytest.mark.vcr()
+    async def test_close_stops_loop_thread(self, temp_db_path):
+        """close() tears down the background loop thread and is idempotent."""
+        config = AppConfig()
+        async with HaikuRAG(temp_db_path, create=True) as client:
+            doc = await client.create_document(
+                content="Foxes and dogs.",
+                uri="test://doc",
+                title="Doc",
+            )
+        assert doc.id
+
+        context = AnalysisContext()
+        sb = Sandbox(db_path=temp_db_path, config=config, context=context)
+        result = await sb.execute(
+            "from pathlib import Path\n"
+            f"print(Path('/documents/{doc.id}/content.txt').read_text())"
+        )
+        assert result.success
+        thread = sb._loop_thread
+        assert thread is not None and thread.is_alive()
+        sb.close()
+        assert not thread.is_alive()
+        sb.close()