deer-flow/backend/app/gateway/upload_ingestion.py
Hyeonsang Cho 29d285731b
fix(uploads): convert the bytes we wrote, not the name they landed under (#5611)
* fix(uploads): convert the bytes we wrote, not the name they landed under

Document conversion re-opened the upload by name after it was already
visible in the thread's uploads directory: the Gateway converted the
committed file_path, and DeerFlowClient converted the copy it had just
placed there. That directory is writable from local and AIO sandboxes,
so a process watching it can replace the name with a symlink in the
window between the upload landing and the converter opening it. The
converter then reads whatever host file the link points at and writes
that content back into the thread as the .md companion, which the
sandbox can read. Reproduced end to end on both paths with a real xlsx:
the companion came back holding the host file's rows.

The Gateway now duplicates the descriptor of the staged file before the
link-commit, copies those bytes into a private directory outside the
uploads tree, and converts there. A descriptor cannot be redirected by
replacing a name, so the conversion input is the content this request
wrote. The client converts the caller's own source file instead of the
copy in uploads; the source is the file the caller handed in, which the
sandbox cannot reach.

Both already wrote the companion without following a symlink, so only
the read side changes. The uploads directory still receives exactly the
same files.

* docs(changelog): note upload conversion source fix (#5611)

* fix(uploads): close the conversion descriptor when staging its copy fails

Review follow-up. The private directory for the conversion copy was
created before the try that owns the duplicated descriptor, so a failure
there — a full or unwritable temporary filesystem — propagated without
closing it. The upload's own cleanup only unlinks the committed name and
releases the sandbox lease, so the descriptor stayed open for the life of
the process and kept the unlinked staged bytes allocated with it; repeated
failures accumulated both.

Directory creation now happens inside that try, and the finally removes
the directory only once it exists.

* fix(uploads): keep the conversion descriptor owned across cancellation

Review follow-up. run_file_io cannot interrupt its worker, so cancelling
the await around os.dup only abandoned the result: the duplicate was
created moments later with nothing left to close it, and it pinned the
staged bytes of an upload whose name the cleanup had already unlinked.
Cancellation after the duplication was just as leaky, because the
commit-path handler caught Exception and CancelledError is not one.

The duplication now runs as its own task, shielded from the caller's
cancellation, and closes its own result when the caller is gone by the
time the worker finishes. The commit path catches BaseException, closing
the descriptor it already owns before re-raising.

Both windows are pinned: one test stalls the duplication worker after it
allocates and cancels ingestion, the other stalls the commit so the
cancellation lands while the descriptor is owned.

* fix(uploads): drain the conversion copy so its descriptor always closes

Review follow-up. The copy worker owns the duplicated descriptor and
closes it in its own finally, but a bare await let a cancellation cancel
the executor job while it was still queued: the worker never ran, so that
finally never ran either, and the enclosing scope had already handed
ownership away and saw None. Draining also keeps a late worker from
writing into a private directory this scope has since removed.

The copy now goes through await_drained, the shield-and-drain helper the
Gateway already uses for offloads that must not be abandoned mid-flight.
Pinned by a test that holds the copy job queued, cancels ingestion, then
releases it and requires the descriptor to come back closed.
2026-09-21 07:34:09 +08:00

408 lines
20 KiB
Python

"""Shared thread-upload ingestion service (Projects Phase 2 Slice C, §7.3 item 3).
Owns the whole pipeline for files landing in a thread's uploads directory:
staging, ``claim_unique_filename``, size checks, optional conversion under
``uploads.auto_convert_documents``, sandbox-readable permissions, and
non-mounted-provider synchronization of original + derived files through the
authorized sandbox request lease. A denied ``sandbox:execute`` retains the
host upload without allocating a sandbox (ordinary uploads behavior);
acquisition, sync and conversion failures follow the ordinary upload error
and cleanup behavior.
Two callers share this pipeline:
- the ordinary uploads endpoint (``routers/uploads.py``), a thin adapter
mapping ``UploadFile`` parts to chunk streams with zero behavior change;
- the project-shelf attach route (``routers/project_documents.py``), which
feeds a staged shelf copy through the same lifecycle (§7.3 item 3).
Behavior-preservation bridge: the pre-extraction upload tests pin this
pipeline by patching collaborators on the ``routers.uploads`` module
namespace (``ensure_uploads_dir``, ``get_sandbox_provider``,
``_get_upload_limits``, ``_auto_convert_documents_enabled``,
``convert_file_to_markdown``, the low-level file machinery). The service
therefore resolves every collaborator through that module object at call
time (:func:`_uploads`) instead of importing the names directly, so those
patches keep binding to the one pipeline both callers use.
"""
from __future__ import annotations
import asyncio
import logging
import os
import shutil
import tempfile
import uuid
from collections.abc import AsyncIterator
from pathlib import Path
from typing import TYPE_CHECKING, Any
from fastapi import HTTPException
from deerflow.config.app_config import AppConfig
from deerflow.utils.file_io import await_drained, run_file_io
if TYPE_CHECKING:
from fastapi import Request
from app.gateway.authz import SandboxRequestLease
from app.gateway.routers.uploads import UploadLimits
logger = logging.getLogger(__name__)
def _close_fd(fd: int | None) -> None:
"""Close a conversion-source descriptor, tolerating an already-closed one."""
if fd is None:
return
try:
os.close(fd)
except OSError:
logger.warning("Failed to close upload conversion descriptor", exc_info=True)
def _close_abandoned_fd(duplication: asyncio.Future) -> None:
"""Close a descriptor whose owner was cancelled before it could take it."""
if duplication.cancelled() or duplication.exception() is not None:
return
_close_fd(duplication.result())
async def _dup_for_conversion(fileno: int) -> int:
"""Duplicate *fileno* off-thread so cancellation cannot strand the copy.
``run_file_io`` cannot interrupt its worker: cancelling the await only
abandons the result, and here that result is an open descriptor nothing
would ever close — repeated cancellations would exhaust the Gateway's
descriptor limit and pin the staged bytes of every unlinked upload. The
duplication therefore runs as its own task, shielded from the caller's
cancellation, and closes itself if the caller is gone by the time the
worker finishes.
"""
duplication = asyncio.ensure_future(run_file_io(os.dup, fileno))
try:
return await asyncio.shield(duplication)
except BaseException:
duplication.add_done_callback(_close_abandoned_fd)
raise
def _copy_fd_to_path(fd: int, dest: Path) -> None:
"""Copy the bytes behind *fd* to *dest*, then close *fd*.
Reads through the descriptor, not the name it was committed under, so the
copy is the content this request wrote even if the name has since been
replaced.
"""
try:
os.lseek(fd, 0, os.SEEK_SET)
with open(dest, "wb") as out:
while chunk := os.read(fd, 1 << 20):
out.write(chunk)
finally:
_close_fd(fd)
def _uploads() -> Any:
"""Return the uploads router module (late binding — see module docstring)."""
from app.gateway.routers import uploads
return uploads
class UnsafeFilenameError(ValueError):
"""The display filename could not be normalized/claimed; ordinary behavior skips it."""
class UnsafeUploadDestinationError(Exception):
"""The claimed destination failed safety validation; carries the claimed name.
Ordinary uploads record the file as skipped (``skipped_files``) and keep
ingesting the rest of the request; attach maps it to an ingestion
failure. Wraps the manager's ``UnsafeUploadPathError`` so the claimed
(deduplicated) filename survives to the caller.
"""
def __init__(self, filename: str) -> None:
super().__init__(f"Unsafe upload destination: {filename}")
self.filename = filename
class ThreadUploadIngestionService:
"""One thread's ingestion session: staging → conversion → permissions → sync.
Lifecycle: :meth:`open` (uploads dir + existing-name seed + optional
sandbox lease), one or
more :meth:`ingest_chunks` calls, :meth:`finalize` (sandbox-readable
permissions + non-mounted provider sync), then :meth:`aclose` (lease
release) — use it as an async context manager. :meth:`cleanup_written`
removes every file written by this session; the ordinary endpoint invokes
it when a per-file failure aborts the request, and attach mirrors that.
"""
def __init__(self, *, request: Request | None, thread_id: str, user_id: str, app_config: AppConfig) -> None:
self._request = request
self._thread_id = thread_id
self._user_id = user_id
self._config = app_config
self._limits: UploadLimits | None = None
self._uploads_dir: Path | None = None
self._sync_to_sandbox = False
self._sandbox_lease: SandboxRequestLease | None = None
self._sandbox: Any = None
self._auto_convert = False
self._seen_filenames: set[str] = set()
self._written_paths: list[Path] = []
self._sync_targets: list[tuple[Path, str]] = []
self._total_size = 0
async def __aenter__(self) -> ThreadUploadIngestionService:
await self.open()
return self
async def __aexit__(self, exc_type: Any, exc: Any, tb: Any) -> None:
await self.aclose()
@property
def limits(self) -> UploadLimits:
assert self._limits is not None, "open() must run before limits is read"
return self._limits
async def open(self) -> None:
"""Resolve limits, ensure the uploads dir, seed existing names, acquire the lease.
Mirrors the ordinary endpoint's setup order: a role denied
``sandbox:execute`` skips acquisition entirely (the host upload still
succeeds); an allowed caller whose sandbox vanishes right after
acquiring is a 500. ``ValueError`` from the uploads-dir resolution
(unsafe thread id) propagates for the caller to map. The existing-name
seed makes every later claim unique against the thread's CURRENT
files, so no ingestion — ordinary upload or shelf attach — ever
silently replaces a conversation file (same-name re-uploads land as
``name_1.ext``); the directory scan is offloaded per the filesystem
convention.
"""
uploads = _uploads()
self._limits = uploads._get_upload_limits(self._config)
self._uploads_dir = await run_file_io(uploads.ensure_uploads_dir, self._thread_id, user_id=self._user_id)
listing = await run_file_io(uploads.list_files_in_dir, self._uploads_dir)
self._seen_filenames.update(entry["filename"] for entry in listing["files"])
sandbox_provider = uploads.get_sandbox_provider()
self._sync_to_sandbox = not uploads._uses_thread_data_mounts(sandbox_provider)
if self._sync_to_sandbox:
self._sandbox_lease = await uploads.try_acquire_sandbox_for_request(
self._request,
sandbox_provider,
self._thread_id,
user_id=self._user_id,
app_config=self._config,
owner_prefix="gateway:upload",
release_on_last=False,
)
self._sandbox = self._sandbox_lease.sandbox
if not self._sandbox_lease.denied and self._sandbox is None:
raise HTTPException(status_code=500, detail="Failed to acquire sandbox")
self._auto_convert = uploads._auto_convert_documents_enabled(self._config)
async def _link_commit_with_retry(self, uploads: Any, staged_path: Path, claimed_name: str) -> tuple[str, Path]:
"""Publish a staged file under *claimed_name*, atomically and with no overwrite.
``os.link`` (inside ``_commit_upload_temp_no_overwrite`` /
``_link_staged_no_overwrite``) fails with :class:`FileExistsError`
when a concurrent session won the name after this session's seed —
the next suffix is claimed against the seeded set and the link
retried, so two concurrent ingestions of one name always land as
``name.ext`` + ``name_1.ext`` with both byte streams intact. The
staged bytes stay hidden under the ``.upload-*.part`` pattern until
the link makes the final name visible, fully formed.
"""
assert self._uploads_dir is not None, "open() must run before committing destinations"
name = claimed_name
while True:
try:
return name, await run_file_io(uploads._link_staged_no_overwrite, staged_path, self._uploads_dir, name)
except FileExistsError:
name = uploads.claim_unique_filename(name, self._seen_filenames)
async def ingest_chunks(self, chunks: AsyncIterator[bytes], *, display_name: str) -> dict[str, Any]:
"""Ingest one file from a chunk stream; return its wire metadata dict.
Owns the per-file pipeline exactly as the ordinary endpoint defines
it: normalize + claim a unique name, staged write with single-file
and request-total size caps (413), an atomic no-overwrite link commit
with next-suffix retry, and conversion with its companion-name claim
when ``uploads.auto_convert_documents`` is on. Raises
:class:`UnsafeFilenameError` (caller logs and skips),
:class:`UnsafeUploadDestinationError` (caller records a skipped
file), ``HTTPException`` (size/over-limit — caller cleans up and
re-raises) or an unexpected error (caller cleans up and answers 500).
"""
assert self._uploads_dir is not None and self._limits is not None, "open() must run before ingest_chunks()"
uploads = _uploads()
try:
original_filename = uploads.normalize_filename(display_name)
except ValueError as exc:
raise UnsafeFilenameError(str(exc)) from exc
safe_filename = uploads.claim_unique_filename(original_filename, self._seen_filenames)
file_size = 0
upload_temp = None
convert_source_fd: int | None = None
try:
upload_temp = await run_file_io(uploads._prepare_upload_destination, self._uploads_dir, safe_filename)
async for chunk in chunks:
file_size += len(chunk)
self._total_size += len(chunk)
if file_size > self._limits.max_file_size:
raise HTTPException(status_code=413, detail=f"File too large: {safe_filename}")
if self._total_size > self._limits.max_total_size:
raise HTTPException(status_code=413, detail="Total upload size too large")
await run_file_io(uploads._write_upload_chunk, upload_temp, chunk)
if self._auto_convert and Path(safe_filename).suffix.lower() in uploads.CONVERTIBLE_EXTENSIONS:
# Conversion must read the bytes this request staged. Once the
# name is committed a sandbox process can replace it with a
# symlink, and converting by name would then pull a host file
# into this thread's uploads. A descriptor on the staged inode
# cannot be redirected that way.
convert_source_fd = await _dup_for_conversion(upload_temp.handle.fileno())
# Link-commit with collision retry: the FileExistsError arm
# leaves the staged part in place for the retry under the next
# suffix (the handle's second close is idempotent).
while True:
try:
file_path = await run_file_io(uploads._commit_upload_temp_no_overwrite, upload_temp, self._uploads_dir, safe_filename)
break
except FileExistsError:
safe_filename = uploads.claim_unique_filename(safe_filename, self._seen_filenames)
upload_temp = None
except uploads.UnsafeUploadPathError as exc:
_close_fd(convert_source_fd)
if upload_temp is not None:
await run_file_io(uploads._abort_upload_temp, upload_temp)
raise UnsafeUploadDestinationError(safe_filename) from exc
except BaseException:
# BaseException, not Exception: a cancellation between the
# duplication and the commit would otherwise leave the descriptor
# open, and nothing downstream owns it yet.
_close_fd(convert_source_fd)
convert_source_fd = None
if upload_temp is not None:
await run_file_io(uploads._abort_upload_temp, upload_temp)
raise
self._written_paths.append(file_path)
virtual_path = uploads.upload_virtual_path(safe_filename)
if self._sync_to_sandbox:
self._sync_targets.append((file_path, virtual_path))
file_info: dict[str, Any] = {
"filename": safe_filename,
"size": file_size,
"path": str(self._uploads_dir / safe_filename),
"virtual_path": virtual_path,
"artifact_url": uploads.upload_artifact_url(self._thread_id, safe_filename),
}
if safe_filename != original_filename:
file_info["original_filename"] = original_filename
logger.info(f"Saved file: {safe_filename} ({file_size} bytes) to {file_info['path']}")
if convert_source_fd is not None:
# The companion gets the same atomic no-overwrite commit as the
# original: staged under the hidden .part pattern, link-committed
# with next-suffix retry — conversion can never silently truncate
# another uploaded or derived file, in this session or a
# concurrent one.
provisional_md_name = Path(safe_filename).with_suffix(".md").name
unique_md_name = uploads.claim_unique_filename(provisional_md_name, self._seen_filenames)
md_staging = self._uploads_dir / f"{uploads.UPLOAD_STAGING_PREFIX}{uuid.uuid4().hex}{uploads.UPLOAD_STAGING_SUFFIX}"
# The staged bytes are copied out of the sandbox-writable tree and
# converted there; the uploads dir only ever receives the result.
# Creating that directory belongs inside the cleanup scope: it can
# fail on its own (a full or unwritable temporary filesystem), and
# the descriptor is already owned here — leaking it would also hold
# the unlinked staged bytes until the process exits.
private_dir: Path | None = None
try:
private_dir = Path(await run_file_io(tempfile.mkdtemp, "-deerflow-convert"))
conversion_source = private_dir / safe_filename
# Hand the descriptor over before the call: the copy closes it
# even when it fails, so this scope must not close it again and
# risk closing an unrelated descriptor that reused the number.
# await_drained, not a bare await: a cancelled await would
# cancel the queued executor job before its worker — and its
# closing finally — ever ran, and draining also keeps the
# worker from writing into a private directory this scope has
# already removed.
staged_fd, convert_source_fd = convert_source_fd, None
await await_drained(run_file_io(_copy_fd_to_path, staged_fd, conversion_source))
md_staged = await uploads.convert_file_to_markdown(conversion_source, output_path=md_staging)
except Exception:
self._seen_filenames.discard(unique_md_name)
await run_file_io(md_staging.unlink, True)
raise
finally:
_close_fd(convert_source_fd)
if private_dir is not None:
await run_file_io(shutil.rmtree, private_dir, True)
if not md_staged:
# Conversion failed and wrote nothing (or a partial staged
# file, removed here): release the claim; holding it would
# rename a later same-stem upload against a name nothing
# occupies.
self._seen_filenames.discard(unique_md_name)
await run_file_io(md_staging.unlink, True)
else:
unique_md_name, md_path = await self._link_commit_with_retry(uploads, Path(md_staged), unique_md_name)
self._written_paths.append(md_path)
md_virtual_path = uploads.upload_virtual_path(md_path.name)
if self._sync_to_sandbox:
self._sync_targets.append((md_path, md_virtual_path))
file_info["markdown_file"] = md_path.name
file_info["markdown_path"] = str(self._uploads_dir / md_path.name)
file_info["markdown_virtual_path"] = md_virtual_path
file_info["markdown_artifact_url"] = uploads.upload_artifact_url(self._thread_id, md_path.name)
return file_info
async def finalize(self) -> None:
"""Make written files sandbox-readable, then sync to a non-mounted sandbox.
Runs after every file of the session ingested successfully — exactly
the ordinary endpoint's tail. Failures here propagate (the host files
stay in place; the caller does not clean them up, matching ordinary
upload behavior for sync-phase errors).
"""
uploads = _uploads()
# Uploaded files are created with 0o600 permissions (owner read/write
# only). In Docker sandbox deployments the gateway writes as root but
# the sandbox process runs as a non-root user (typically UID 1000).
# Without group/other read bits the sandbox cannot access the files —
# whether the uploads directory is bind-mounted into the container or
# synced via sandbox.update_file. Always add group/other read bits so
# every sandbox configuration can read the uploaded content.
await run_file_io(uploads._make_uploaded_paths_sandbox_readable, self._written_paths)
if self._sync_to_sandbox and self._sandbox is not None:
for file_path, virtual_path in self._sync_targets:
await run_file_io(uploads._sync_upload_to_sandbox, self._sandbox, file_path, virtual_path)
async def cleanup_written(self) -> None:
"""Remove every file this session wrote (ordinary rejected-request cleanup)."""
if not self._written_paths:
return
uploads = _uploads()
await run_file_io(uploads._cleanup_uploaded_paths, self._written_paths)
self._written_paths = []
async def aclose(self) -> None:
"""Release the sandbox request lease (failures are logged, never raised)."""
if self._sandbox_lease is None:
return
try:
await self._sandbox_lease.release()
except Exception:
logger.warning(
"Failed to release sandbox request lease after upload sync: %s",
self._sandbox_lease.sandbox_id,
exc_info=True,
)