95fdb46c3a
analysis.code_timeout was enforced only in _run_on_loop, the bridge for database-bound reads; metadata.json, the cached JSONL files and in-code search() and list_documents() never looked at the clock, and Monty's watchdog counts compute only. Every host call now checks the deadline before it starts. Host errors keep their message for every caller: the masking added for the MCP server goes, and the sandbox is one path for the capability and the server alike.
704 lines
27 KiB
Python
704 lines
27 KiB
Python
import asyncio
|
|
import json
|
|
import os
|
|
from collections.abc import AsyncIterator, Callable, Coroutine
|
|
from contextlib import asynccontextmanager, suppress
|
|
from dataclasses import dataclass
|
|
from itertools import zip_longest
|
|
from typing import TYPE_CHECKING, Any, Literal
|
|
|
|
import pydantic_monty
|
|
from pydantic_monty import (
|
|
AsyncMonty,
|
|
AsyncMontySession,
|
|
CallbackFile,
|
|
OSAccess,
|
|
ResourceLimits,
|
|
)
|
|
|
|
from haiku.rag.config.models import AppConfig
|
|
from haiku.rag.context import build_toc
|
|
from haiku.rag.sandbox.dependencies import AnalysisContext
|
|
from haiku.rag.store.models.chunk import Chunk, SearchResult
|
|
from haiku.rag.store.models.document_item import PICTURE_REF_PREFIX, DocumentItem
|
|
from haiku.rag.utils import gather_all
|
|
|
|
if TYPE_CHECKING:
|
|
from pathlib import Path, PurePosixPath
|
|
|
|
from haiku.rag.client import HaikuRAG
|
|
from haiku.rag.client.scope import DatabaseScope
|
|
|
|
|
|
_MAX_HOST_CALLS = 10_000_000
|
|
|
|
|
|
@dataclass
|
|
class SandboxResult:
|
|
"""Result of executing code in the sandbox."""
|
|
|
|
stdout: str
|
|
stderr: str
|
|
success: bool
|
|
|
|
|
|
def recovery_hint(stderr: str) -> str:
|
|
"""Name the workaround for sandbox limits models trip over repeatedly.
|
|
|
|
The instructions already say file objects are not iterable, and models write
|
|
``for line in open(...)`` regardless. Carrying the fix in the error gives
|
|
them something to act on for the retry.
|
|
"""
|
|
if "TextIOWrapper" in stderr and "not iterable" in stderr:
|
|
return (
|
|
"\n\nHint: file objects cannot be iterated here. Read lines with "
|
|
'.readlines() or .read().split("\\n").'
|
|
)
|
|
return ""
|
|
|
|
|
|
class Sandbox:
|
|
"""Execute code in a sandboxed Python interpreter.
|
|
|
|
Uses pydantic-monty, a minimal secure Python interpreter written in Rust.
|
|
The interpreter runs in a subprocess worker checked out of an ``AsyncMonty``
|
|
pool. External functions (search, list_documents) are called by Monty code
|
|
using ``await`` and resolved asynchronously on the host. Documents are
|
|
exposed via a virtual filesystem at ``/documents/{id}/``: ``metadata.json``,
|
|
``content.txt``, ``items.jsonl``, ``chunks.jsonl`` and ``toc.json``.
|
|
|
|
The session persists across ``execute()`` calls within the same Sandbox
|
|
instance — variables carry over. Call ``close()`` to return the worker to
|
|
the pool, shut the pool down and release any held connection.
|
|
|
|
sandbox = Sandbox(db_path, config, context)
|
|
result = await sandbox.execute("x = await search('query')")
|
|
result = await sandbox.execute("print(x[0]['content'])") # x persists
|
|
await sandbox.close()
|
|
|
|
All database access runs on the event loop that drives ``execute()``. Monty's
|
|
file callbacks are synchronous and run off that loop while ``feed_run`` is
|
|
awaited, so they bridge back to it via ``run_coroutine_threadsafe`` without
|
|
deadlocking. When a ``rag`` connection is supplied it is used for every read,
|
|
so an analysis run drives a single connection on a single loop. Otherwise a
|
|
scope covering several databases opens a federated client once and holds it
|
|
until ``close()``, and a single database is opened per read.
|
|
"""
|
|
|
|
_scope: "DatabaseScope"
|
|
_config: AppConfig
|
|
_context: AnalysisContext
|
|
_rag: "HaikuRAG | None"
|
|
_owners: dict[str, "HaikuRAG"]
|
|
_lock: "asyncio.Lock | None"
|
|
_search_results: "list[SearchResult]"
|
|
_doc_items: dict[str, list["DocumentItem"]]
|
|
_doc_chunk_index: dict[str, dict[str, list[str]]]
|
|
_items_jsonl_cache: dict[str, str]
|
|
_chunks_jsonl_cache: dict[str, str]
|
|
_toc_json_cache: dict[str, str]
|
|
_opened: "HaikuRAG | None"
|
|
_pool: AsyncMonty | None
|
|
_session: AsyncMontySession | None
|
|
_vfs: OSAccess | None
|
|
_loop: asyncio.AbstractEventLoop | None
|
|
_deadline: float | None
|
|
|
|
def __init__(
|
|
self,
|
|
db_path: "Path | str | None",
|
|
config: AppConfig,
|
|
context: AnalysisContext,
|
|
rag: "HaikuRAG | None" = None,
|
|
lock: "asyncio.Lock | None" = None,
|
|
):
|
|
from haiku.rag.client.scope import DatabaseScope
|
|
|
|
self._configure(
|
|
DatabaseScope.resolve(config, database_path=db_path),
|
|
config,
|
|
context,
|
|
rag,
|
|
lock,
|
|
)
|
|
|
|
@classmethod
|
|
def _covering(
|
|
cls,
|
|
scope: "DatabaseScope",
|
|
config: AppConfig,
|
|
context: AnalysisContext,
|
|
rag: "HaikuRAG | None" = None,
|
|
lock: "asyncio.Lock | None" = None,
|
|
) -> "Sandbox":
|
|
"""A sandbox over databases someone already resolved.
|
|
|
|
Internal: the public constructor takes a path and resolves it, which is
|
|
its own job. This is for callers that did the resolving, as
|
|
``HaikuRAG._covering`` is. It bypasses ``__init__``: the scope it is
|
|
handed is the only one resolved.
|
|
"""
|
|
sandbox = cls.__new__(cls)
|
|
sandbox._configure(scope, config, context, rag, lock)
|
|
return sandbox
|
|
|
|
def _configure(
|
|
self,
|
|
scope: "DatabaseScope",
|
|
config: AppConfig,
|
|
context: AnalysisContext,
|
|
rag: "HaikuRAG | None",
|
|
lock: "asyncio.Lock | None",
|
|
) -> None:
|
|
"""The state every sandbox starts with, however its scope was reached."""
|
|
self._scope = scope
|
|
self._config = config
|
|
self._context = context
|
|
self._rag = rag
|
|
self._opened = None
|
|
self._owners = {}
|
|
self._lock = lock
|
|
self._search_results = []
|
|
self._doc_items = {}
|
|
self._doc_chunk_index = {}
|
|
self._items_jsonl_cache = {}
|
|
self._chunks_jsonl_cache = {}
|
|
self._toc_json_cache = {}
|
|
self._pool = None
|
|
self._session = None
|
|
self._vfs = None
|
|
self._loop = None
|
|
self._deadline = None
|
|
|
|
@asynccontextmanager
|
|
async def _connection(
|
|
self, owner: "HaikuRAG | None" = None
|
|
) -> "AsyncIterator[HaikuRAG]":
|
|
"""Yield the shared connection, or an ephemeral read-only one.
|
|
|
|
The lock, where there is one, is held for the block: what a caller does
|
|
to the shared connection is serialized against the capability's own tool
|
|
calls, since both hold that client. It guards that connection and not
|
|
the databases underneath, which have their own locks.
|
|
|
|
`owner` is the client holding one document, for the reads addressed to a
|
|
single document. The shared connection covers a set of databases and has
|
|
no repositories of its own, so those reads have to name their owner. An
|
|
owner is a session of its own, so it is yielded unserialized.
|
|
"""
|
|
if owner is not None:
|
|
yield owner
|
|
return
|
|
if self._rag is not None:
|
|
if self._lock is not None:
|
|
async with self._lock:
|
|
yield self._rag
|
|
else:
|
|
yield self._rag
|
|
return
|
|
if self._scope.covers_multiple:
|
|
yield await self._open_connection()
|
|
return
|
|
from haiku.rag.client import HaikuRAG
|
|
|
|
async with HaikuRAG._covering(self._scope, self._config, read_only=True) as rag:
|
|
yield rag
|
|
|
|
async def _open_connection(self) -> "HaikuRAG":
|
|
"""Open and retain a federated client for owner-backed reads."""
|
|
if self._opened is None:
|
|
from haiku.rag.client import HaikuRAG
|
|
|
|
self._opened = HaikuRAG._covering(self._scope, self._config, read_only=True)
|
|
await self._opened.__aenter__()
|
|
return self._opened
|
|
|
|
async def _documents(self) -> "tuple[list[Any], dict[str, HaikuRAG]]":
|
|
"""Every document in scope, and the client holding each of them.
|
|
|
|
Owners are empty where one connection serves every read. The selection
|
|
resolves as a search resolves it, so a database the question excluded
|
|
cannot be mounted.
|
|
"""
|
|
async with self._connection() as rag:
|
|
if not rag.covers_multiple:
|
|
if not await rag.clients_covering(self._context.sources):
|
|
return [], {}
|
|
docs = await rag.list_documents(filter=self._context.filter)
|
|
return docs, {}
|
|
owners = await rag.clients_covering(self._context.sources)
|
|
# Resolve owners under the shared-client lock; owner sessions perform
|
|
# reads independently.
|
|
groups = await gather_all(
|
|
*(owner.list_documents(filter=self._context.filter) for owner in owners)
|
|
)
|
|
# Interleaved: a truncated listing still shows every database.
|
|
docs = [doc for row in zip_longest(*groups) for doc in row if doc is not None]
|
|
return docs, self._holders(owners, groups)
|
|
|
|
@staticmethod
|
|
def _holders(
|
|
owners: "list[HaikuRAG]", groups: "list[list[Any]]"
|
|
) -> "dict[str, HaikuRAG]":
|
|
"""Map each document id to the database holding it, rejecting an id two
|
|
databases claim. The mount has one `/documents/{id}/` path per id.
|
|
"""
|
|
holders: dict[str, HaikuRAG] = {}
|
|
held_by: dict[str, str | None] = {}
|
|
for owner, group in zip(owners, groups, strict=True):
|
|
for doc in group:
|
|
if not doc.id: # pragma: no cover - stored rows always carry an id
|
|
continue
|
|
if doc.id in holders:
|
|
raise ValueError(
|
|
f"document {doc.id} is in databases {held_by[doc.id]!r} and "
|
|
f"{owner.source!r}; analysis mounts one document per id"
|
|
)
|
|
holders[doc.id] = owner
|
|
held_by[doc.id] = owner.source
|
|
return holders
|
|
|
|
def _run_on_loop(self, coro: Coroutine[Any, Any, Any]) -> Any:
|
|
"""Run a coroutine on the execute() loop from a synchronous callback.
|
|
|
|
Called off the loop while ``feed_run`` is awaited, so scheduling onto it
|
|
and blocking for the result is safe.
|
|
|
|
Blocking here suspends the worker, and Monty checks its duration budget
|
|
between interpreter steps, so it cannot check while a read is in flight.
|
|
Enforce the budget before starting another read, or code that reads in a
|
|
loop overruns it by however long the outstanding reads take. Raising from
|
|
inside the callback answers the worker's suspension, which keeps the
|
|
session usable — cancelling ``feed_run`` from outside does not, and wedges
|
|
the protocol.
|
|
"""
|
|
assert self._loop is not None, (
|
|
"VFS reads happen during execute(); the loop must be captured first."
|
|
)
|
|
if self._past_deadline():
|
|
coro.close()
|
|
raise self._time_limit()
|
|
return asyncio.run_coroutine_threadsafe(coro, self._loop).result()
|
|
|
|
def _past_deadline(self) -> bool:
|
|
return (
|
|
self._deadline is not None
|
|
and self._loop is not None
|
|
and self._loop.time() > self._deadline
|
|
)
|
|
|
|
def _time_limit(self) -> TimeoutError:
|
|
return TimeoutError(
|
|
"time limit exceeded: no further document reads or calls after "
|
|
f"{self._config.analysis.code_timeout}s"
|
|
)
|
|
|
|
def _check_deadline(self) -> None:
|
|
"""Refuse a host call once the call's time is up.
|
|
|
|
Monty's watchdog counts only time the worker spends computing, so every
|
|
host call, a file served from memory and an in-code search included,
|
|
checks the deadline before it runs.
|
|
"""
|
|
if self._past_deadline():
|
|
raise self._time_limit()
|
|
|
|
def _timed(
|
|
self, read: Callable[["PurePosixPath"], str]
|
|
) -> Callable[["PurePosixPath"], str]:
|
|
def call(path: "PurePosixPath") -> str:
|
|
self._check_deadline()
|
|
return read(path)
|
|
|
|
return call
|
|
|
|
async def _discard_session(self) -> None:
|
|
"""Drop a session whose worker is gone.
|
|
|
|
The session object is unusable once its worker dies: it answers every
|
|
later call with ``RuntimeError: this checkout has already been
|
|
finished``. Clearing it makes ``_ensure_initialized`` check out a
|
|
replacement, at the cost of the variables the dead worker held.
|
|
"""
|
|
session, self._session = self._session, None
|
|
if session is not None:
|
|
with suppress(Exception):
|
|
await session.__aexit__(None, None, None)
|
|
|
|
async def close(self) -> None:
|
|
"""Return the worker to the pool, shut the pool down and release any
|
|
held connection. Idempotent, and each release happens whatever the
|
|
others raise."""
|
|
if self._session is not None:
|
|
session, self._session = self._session, None
|
|
with suppress(Exception):
|
|
await session.__aexit__(None, None, None)
|
|
if self._pool is not None:
|
|
pool, self._pool = self._pool, None
|
|
with suppress(Exception):
|
|
await pool.__aexit__(None, None, None)
|
|
if self._opened is not None:
|
|
opened, self._opened = self._opened, None
|
|
with suppress(Exception):
|
|
await opened.__aexit__(None, None, None)
|
|
|
|
def _build_external_functions(self) -> dict[str, Any]:
|
|
"""Build async external functions for the Monty interpreter."""
|
|
context = self._context
|
|
|
|
async def search(query: str, limit: int = 10) -> list[dict[str, Any]]:
|
|
self._check_deadline()
|
|
# Picture bytes are deliberately not attached to in-code search
|
|
# results: the Monty interpreter has no PIL/base64/hashlib, so the
|
|
# agent's Python can't do anything with them. The driving model
|
|
# gets figures through the top-level `search` tool when the
|
|
# question is visual; in-code search is for structural work.
|
|
async with self._connection() as rag:
|
|
results = await rag.search(
|
|
query,
|
|
limit=limit,
|
|
filter=context.filter,
|
|
sources=context.sources,
|
|
)
|
|
expanded = await rag.expand_context(results)
|
|
self._search_results.extend(expanded)
|
|
out: list[dict[str, Any]] = []
|
|
for r in expanded:
|
|
picture_refs = [
|
|
ref for ref in r.doc_item_refs if ref.startswith(PICTURE_REF_PREFIX)
|
|
]
|
|
out.append(
|
|
{
|
|
"chunk_id": r.chunk_id,
|
|
"content": r.content,
|
|
"source": r.source,
|
|
"document_id": r.document_id,
|
|
"document_title": r.document_title,
|
|
"document_uri": r.document_uri,
|
|
"score": r.score,
|
|
"page_numbers": r.page_numbers,
|
|
"headings": r.headings,
|
|
"doc_item_refs": r.doc_item_refs,
|
|
"labels": r.labels,
|
|
"picture_refs": picture_refs,
|
|
"chunk_meta": r.chunk_meta,
|
|
}
|
|
)
|
|
return out
|
|
|
|
async def list_documents() -> list[dict[str, Any]]:
|
|
self._check_deadline()
|
|
docs, _ = await self._documents()
|
|
return [
|
|
{
|
|
"id": d.id,
|
|
"title": d.title,
|
|
"uri": d.uri,
|
|
"created_at": str(d.created_at),
|
|
"source": d.source,
|
|
"metadata": d.metadata,
|
|
}
|
|
for d in docs
|
|
]
|
|
|
|
return {
|
|
"search": search,
|
|
"list_documents": list_documents,
|
|
}
|
|
|
|
async def _build_vfs(self) -> OSAccess:
|
|
"""Build the virtual filesystem with document data.
|
|
|
|
Mounts per-document directories with:
|
|
- metadata.json: CallbackFile (eager, small)
|
|
- content.txt: CallbackFile (lazy, can be large)
|
|
- items.jsonl: CallbackFile (lazy, bulk-cached)
|
|
- chunks.jsonl: CallbackFile (lazy, bulk-cached)
|
|
- toc.json: CallbackFile (lazy, bulk-cached)
|
|
"""
|
|
files: list[CallbackFile] = []
|
|
|
|
def _deny_write(_path: "PurePosixPath", _content: str | bytes) -> None:
|
|
raise PermissionError(f"Document files are read-only: {_path}")
|
|
|
|
docs, self._owners = await self._documents()
|
|
|
|
doc_titles = {doc.id: doc.title for doc in docs if doc.id}
|
|
|
|
sandbox = self
|
|
|
|
def _get_items(did: str) -> list[DocumentItem]:
|
|
"""Fetch items for one doc, cached on the sandbox."""
|
|
cached = sandbox._doc_items.get(did)
|
|
if cached is not None:
|
|
return cached
|
|
|
|
async def _fetch() -> list[DocumentItem]:
|
|
async with sandbox._connection(sandbox._owners.get(did)) as rag:
|
|
return await rag.document_item_repository.get_all_items(did)
|
|
|
|
items = sandbox._run_on_loop(_fetch())
|
|
sandbox._doc_items[did] = items
|
|
return items
|
|
|
|
def _get_chunk_index(did: str) -> dict[str, list[str]]:
|
|
"""Fetch the self_ref → chunk_ids index for one doc, cached."""
|
|
cached = sandbox._doc_chunk_index.get(did)
|
|
if cached is not None:
|
|
return cached
|
|
|
|
async def _fetch() -> dict[str, list[str]]:
|
|
async with sandbox._connection(sandbox._owners.get(did)) as rag:
|
|
index = (
|
|
await rag.chunk_repository.get_chunk_ids_by_self_ref_grouped(
|
|
[did]
|
|
)
|
|
)
|
|
return index.get(did, {})
|
|
|
|
chunk_index = sandbox._run_on_loop(_fetch())
|
|
sandbox._doc_chunk_index[did] = chunk_index
|
|
return chunk_index
|
|
|
|
def _make_items_reader(
|
|
did: str,
|
|
) -> Callable[["PurePosixPath"], str]:
|
|
def read_items(_path: "PurePosixPath") -> str:
|
|
cached = sandbox._items_jsonl_cache.get(did)
|
|
if cached is not None:
|
|
return cached
|
|
items = _get_items(did)
|
|
chunk_index = _get_chunk_index(did)
|
|
jsonl = "\n".join(
|
|
json.dumps(
|
|
{
|
|
"self_ref": item.self_ref,
|
|
"label": item.label,
|
|
"text": item.text,
|
|
"page_numbers": item.page_numbers,
|
|
"heading_level": item.heading_level,
|
|
"chunk_ids": chunk_index.get(item.self_ref, []),
|
|
},
|
|
ensure_ascii=False,
|
|
)
|
|
for item in items
|
|
)
|
|
sandbox._items_jsonl_cache[did] = jsonl
|
|
return jsonl
|
|
|
|
return read_items
|
|
|
|
def _make_chunks_reader(
|
|
did: str,
|
|
) -> Callable[["PurePosixPath"], str]:
|
|
def read_chunks(_path: "PurePosixPath") -> str:
|
|
cached = sandbox._chunks_jsonl_cache.get(did)
|
|
if cached is not None:
|
|
return cached
|
|
|
|
async def _fetch() -> list[Chunk]:
|
|
async with sandbox._connection(sandbox._owners.get(did)) as rag:
|
|
return await rag.chunk_repository.get_by_document_id(did)
|
|
|
|
chunks = sandbox._run_on_loop(_fetch())
|
|
jsonl = "\n".join(
|
|
json.dumps(
|
|
{"chunk_id": chunk.id, "metadata": chunk.metadata},
|
|
ensure_ascii=False,
|
|
)
|
|
for chunk in chunks
|
|
)
|
|
sandbox._chunks_jsonl_cache[did] = jsonl
|
|
return jsonl
|
|
|
|
return read_chunks
|
|
|
|
def _make_toc_reader(
|
|
did: str,
|
|
) -> Callable[["PurePosixPath"], str]:
|
|
def read_toc(_path: "PurePosixPath") -> str:
|
|
cached = sandbox._toc_json_cache.get(did)
|
|
if cached is not None:
|
|
return cached
|
|
items = _get_items(did)
|
|
chunk_index = _get_chunk_index(did)
|
|
toc = json.dumps(
|
|
{
|
|
"doc_id": did,
|
|
"title": doc_titles.get(did),
|
|
"tree": build_toc(items, chunk_index),
|
|
},
|
|
ensure_ascii=False,
|
|
)
|
|
sandbox._toc_json_cache[did] = toc
|
|
return toc
|
|
|
|
return read_toc
|
|
|
|
for doc in docs:
|
|
if not doc.id: # pragma: no cover - stored rows always carry an id
|
|
continue
|
|
doc_id: str = doc.id
|
|
doc_dir = f"/documents/{doc_id}"
|
|
|
|
metadata = json.dumps(
|
|
{
|
|
"id": doc_id,
|
|
"title": doc.title,
|
|
"uri": doc.uri,
|
|
"created_at": str(doc.created_at),
|
|
"metadata": doc.metadata,
|
|
},
|
|
ensure_ascii=False,
|
|
)
|
|
|
|
# MemoryFile has no write hook, so metadata.json goes through the
|
|
# same read and deny pair as the rest. Its content is already built.
|
|
files.append(
|
|
CallbackFile(
|
|
f"{doc_dir}/metadata.json",
|
|
read=self._timed(lambda _path, text=metadata: text),
|
|
write=_deny_write,
|
|
)
|
|
)
|
|
|
|
def _make_content_reader(
|
|
did: str,
|
|
) -> Callable[["PurePosixPath"], str]:
|
|
def read_content(_path: "PurePosixPath") -> str:
|
|
async def _fetch() -> str:
|
|
async with sandbox._connection(sandbox._owners.get(did)) as rag:
|
|
content = await rag.document_repository.get_content(did)
|
|
return content or ""
|
|
|
|
return sandbox._run_on_loop(_fetch())
|
|
|
|
return read_content
|
|
|
|
files.append(
|
|
CallbackFile(
|
|
f"{doc_dir}/content.txt",
|
|
read=self._timed(_make_content_reader(doc_id)),
|
|
write=_deny_write,
|
|
)
|
|
)
|
|
files.append(
|
|
CallbackFile(
|
|
f"{doc_dir}/items.jsonl",
|
|
read=self._timed(_make_items_reader(doc_id)),
|
|
write=_deny_write,
|
|
)
|
|
)
|
|
files.append(
|
|
CallbackFile(
|
|
f"{doc_dir}/chunks.jsonl",
|
|
read=self._timed(_make_chunks_reader(doc_id)),
|
|
write=_deny_write,
|
|
)
|
|
)
|
|
# HAIKU_RAG_DISABLE_TOC is an evaluation-time toggle for measuring
|
|
# whether toc.json's outline view earns its place in the VFS.
|
|
# Production callers should leave it unset.
|
|
if not os.environ.get("HAIKU_RAG_DISABLE_TOC"):
|
|
files.append(
|
|
CallbackFile(
|
|
f"{doc_dir}/toc.json",
|
|
read=self._timed(_make_toc_reader(doc_id)),
|
|
write=_deny_write,
|
|
)
|
|
)
|
|
|
|
return OSAccess(files)
|
|
|
|
def _session_limits(self) -> ResourceLimits:
|
|
"""Resource limits for the worker session.
|
|
|
|
Monty spends ``max_duration_secs`` across the session's whole life, and
|
|
the session is reused so variables persist between calls: the budget
|
|
covers the whole run. ``code_timeout`` is enforced per call elsewhere: past
|
|
its deadline no further host call starts (``_check_deadline``), and the
|
|
pool's ``request_timeout`` bounds compute.
|
|
|
|
``max_suspensions`` counts host callbacks per session, document reads
|
|
included, defaults to 1000 and cannot be disabled. The time budgets are
|
|
the governors here, so it is set where no program reaches it.
|
|
"""
|
|
analysis = self._config.analysis
|
|
return {
|
|
"max_duration_secs": analysis.code_timeout * analysis.max_executions,
|
|
"max_suspensions": _MAX_HOST_CALLS,
|
|
}
|
|
|
|
async def _ensure_initialized(self) -> tuple[AsyncMontySession, OSAccess]:
|
|
"""Check out a worker session and build the VFS on first use."""
|
|
if self._vfs is None:
|
|
self._vfs = await self._build_vfs()
|
|
if self._pool is None:
|
|
# The watchdog counts only time the worker spends running code, so a
|
|
# read that blocks the worker never trips it. That leaves the two
|
|
# limits disjoint: this one bounds a call that computes, and the read
|
|
# deadline bounds a call that reads.
|
|
pool = AsyncMonty(request_timeout=self._config.analysis.code_timeout)
|
|
await pool.__aenter__()
|
|
self._pool = pool
|
|
if self._session is None:
|
|
session = self._pool.checkout(limits=self._session_limits())
|
|
await session.__aenter__()
|
|
self._session = session
|
|
assert self._session is not None and self._vfs is not None
|
|
return self._session, self._vfs
|
|
|
|
async def execute(self, code: str) -> SandboxResult:
|
|
"""Execute Python code in the Monty worker session.
|
|
|
|
Variables persist across calls within the same Sandbox instance.
|
|
"""
|
|
# Monty's synchronous file callbacks bridge DB reads back to this loop.
|
|
self._loop = asyncio.get_running_loop()
|
|
self._deadline = self._loop.time() + self._config.analysis.code_timeout
|
|
session, vfs = await self._ensure_initialized()
|
|
external_fns = self._build_external_functions()
|
|
|
|
stdout_lines: list[str] = []
|
|
|
|
def print_callback( # pragma: no cover - runs on Monty's worker thread
|
|
_stream: Literal["stdout", "stderr"], text: str
|
|
) -> None:
|
|
stdout_lines.append(text)
|
|
|
|
max_chars = self._config.analysis.max_output_chars
|
|
|
|
try:
|
|
output = await session.feed_run(
|
|
code,
|
|
external_lookup=external_fns,
|
|
print_callback=print_callback,
|
|
os=vfs,
|
|
)
|
|
except (pydantic_monty.MontyError, RuntimeError) as e:
|
|
stdout = "".join(stdout_lines)
|
|
if len(stdout) > max_chars:
|
|
stdout = stdout[:max_chars] + "\n... (output truncated)"
|
|
stderr = str(e)
|
|
# A crash kills the worker, and a protocol error leaves it out of
|
|
# step. Both poison the session. Bad user code does not.
|
|
if isinstance(e, pydantic_monty.MontyCrashedError | RuntimeError):
|
|
await self._discard_session()
|
|
stderr = (
|
|
f"{stderr}\n\nThe interpreter restarted. Variables from "
|
|
"earlier calls are gone."
|
|
)
|
|
return SandboxResult(stdout=stdout, stderr=stderr, success=False)
|
|
|
|
stdout = "".join(stdout_lines)
|
|
if output is not None:
|
|
stdout_with_output = f"{stdout}{output}" if stdout else str(output)
|
|
else:
|
|
stdout_with_output = stdout
|
|
|
|
if len(stdout_with_output) > max_chars:
|
|
stdout_with_output = (
|
|
stdout_with_output[:max_chars] + "\n... (output truncated)"
|
|
)
|
|
|
|
return SandboxResult(stdout=stdout_with_output, stderr="", success=True)
|