Trimming and simpliciation

Editing a document in a way that renames its file returned a response carrying the modified time from before the move. The deferred file move runs after the response is built and bumps modified, so the stored value was newer than the one the client received. The frontend then saw a newer server modified time and showed a spurious "document was updated" prompt whose reload discarded local edits.

DocumentViewSet.update now flushes pending file moves right after saving and before reading the document back, so the response reflects the final filename and modified time.

docs/changelog.md

@@ -1,5 +1,72 @@
 # Changelog
 
+## paperless-ngx 3.4.2
+
+### Bug Fixes
+
+- Fix: mobile save button display obsessing [@shamoon](https://github.com/shamoon) ([#14676](https://github.com/paperless-ngx/paperless-ngx/pull/14676))
+- Fix: make index and llmindex init wait for tesseract language install [@stumpylog](https://github.com/stumpylog) ([#14670](https://github.com/paperless-ngx/paperless-ngx/pull/14670))
+- Fix: prevent loss of save view name on save + close [@shamoon](https://github.com/shamoon) ([#14667](https://github.com/paperless-ngx/paperless-ngx/pull/14667))
+- Fixhancement: doc detail mobile ux tweaks [@shamoon](https://github.com/shamoon) ([#14664](https://github.com/paperless-ngx/paperless-ngx/pull/14664))
+- Fix: Bumps filelock to 4.2.6 to pick up ACL related fixes [@stumpylog](https://github.com/stumpylog) ([#14656](https://github.com/paperless-ngx/paperless-ngx/pull/14656))
+- Fix: ensure import bulk\_edit tasks in Celery workers [@shamoon](https://github.com/shamoon) ([#14653](https://github.com/paperless-ngx/paperless-ngx/pull/14653))
+- Fixhancement: add tag names to document history list [@shamoon](https://github.com/shamoon) ([#14625](https://github.com/paperless-ngx/paperless-ngx/pull/14625))
+- Fix: add button type to zoom buttons to prevent form submission [@shamoon](https://github.com/shamoon) ([#14624](https://github.com/paperless-ngx/paperless-ngx/pull/14624))
+- Fix: move document list preview button to own column [@shamoon](https://github.com/shamoon) ([#14610](https://github.com/paperless-ngx/paperless-ngx/pull/14610))
+- Fix: Various small bugs found during typing updates [@stumpylog](https://github.com/stumpylog) ([#14593](https://github.com/paperless-ngx/paperless-ngx/pull/14593))
+- Fix: fix custom fields dropdown text size [@shamoon](https://github.com/shamoon) ([#14606](https://github.com/paperless-ngx/paperless-ngx/pull/14606))
+
+### Dependencies
+
+<details>
+<summary>14 changes</summary>
+
+- Chore: ditch the abandoned ngx-bootstrap-icons, roll our own [@shamoon](https://github.com/shamoon) ([#14665](https://github.com/paperless-ngx/paperless-ngx/pull/14665))
+- chore(deps): bump turbohtml to 1.15.3 [@stumpylog](https://github.com/stumpylog) ([#14662](https://github.com/paperless-ngx/paperless-ngx/pull/14662))
+- docker-compose(deps): bump gotenberg/gotenberg from 8.37 to 8.38 in /docker/compose @[dependabot[bot]](https://github.com/apps/dependabot) ([#14634](https://github.com/paperless-ngx/paperless-ngx/pull/14634))
+- chore(deps): bump Django, django-guardian, django-filter, drf-spectacular-sidecar [@stumpylog](https://github.com/stumpylog) ([#14661](https://github.com/paperless-ngx/paperless-ngx/pull/14661))
+- docker-compose(deps): bump greenmail/standalone from 2.1.14 to 2.1.15 in /docker/compose @[dependabot[bot]](https://github.com/apps/dependabot) ([#14631](https://github.com/paperless-ngx/paperless-ngx/pull/14631))
+- docker-compose(deps): Bump nginx from 1.31.6-alpine to 1.31.7-alpine in /docker/compose @[dependabot[bot]](https://github.com/apps/dependabot) ([#14633](https://github.com/paperless-ngx/paperless-ngx/pull/14633))
+- Chore(deps): Bump the frontend-angular-dependencies group across 1 directory with 13 updates @[dependabot[bot]](https://github.com/apps/dependabot) ([#14637](https://github.com/paperless-ngx/paperless-ngx/pull/14637))
+- Chore(deps-dev): Bump @playwright/test from 1.62.1 to 1.64.1 in /src-ui @[dependabot[bot]](https://github.com/apps/dependabot) ([#14641](https://github.com/paperless-ngx/paperless-ngx/pull/14641))
+- Chore(deps): Bump pdfjs-dist from 6.3.289 to 6.4.299 in /src-ui @[dependabot[bot]](https://github.com/apps/dependabot) ([#14644](https://github.com/paperless-ngx/paperless-ngx/pull/14644))
+- Chore(deps): Bump uuid from 14.0.2 to 14.0.3 in /src-ui @[dependabot[bot]](https://github.com/apps/dependabot) ([#14640](https://github.com/paperless-ngx/paperless-ngx/pull/14640))
+- Chore(deps-dev): Bump the frontend-eslint-dependencies group across 1 directory with 4 updates @[dependabot[bot]](https://github.com/apps/dependabot) ([#14638](https://github.com/paperless-ngx/paperless-ngx/pull/14638))
+- Chore(deps): Bump marked from 18.0.14 to 18.1.1 in /src-ui @[dependabot[bot]](https://github.com/apps/dependabot) ([#14643](https://github.com/paperless-ngx/paperless-ngx/pull/14643))
+- Chore(deps-dev): Bump prettier from 3.9.9 to 3.9.10 in /src-ui @[dependabot[bot]](https://github.com/apps/dependabot) ([#14645](https://github.com/paperless-ngx/paperless-ngx/pull/14645))
+- Chore(deps-dev): Bump @types/node from 26.6.2 to 26.6.5 in /src-ui @[dependabot[bot]](https://github.com/apps/dependabot) ([#14639](https://github.com/paperless-ngx/paperless-ngx/pull/14639))
+
+</details>
+
+### All App Changes
+
+<details>
+<summary>21 changes</summary>
+
+- Fix: mobile save button display obsessing [@shamoon](https://github.com/shamoon) ([#14676](https://github.com/paperless-ngx/paperless-ngx/pull/14676))
+- Fix: prevent loss of save view name on save + close [@shamoon](https://github.com/shamoon) ([#14667](https://github.com/paperless-ngx/paperless-ngx/pull/14667))
+- Chore: ditch the abandoned ngx-bootstrap-icons, roll our own [@shamoon](https://github.com/shamoon) ([#14665](https://github.com/paperless-ngx/paperless-ngx/pull/14665))
+- Fixhancement: doc detail mobile ux tweaks [@shamoon](https://github.com/shamoon) ([#14664](https://github.com/paperless-ngx/paperless-ngx/pull/14664))
+- chore(deps): bump turbohtml to 1.15.3 [@stumpylog](https://github.com/stumpylog) ([#14662](https://github.com/paperless-ngx/paperless-ngx/pull/14662))
+- chore(deps): bump Django, django-guardian, django-filter, drf-spectacular-sidecar [@stumpylog](https://github.com/stumpylog) ([#14661](https://github.com/paperless-ngx/paperless-ngx/pull/14661))
+- Fix: Bumps filelock to 4.2.6 to pick up ACL related fixes [@stumpylog](https://github.com/stumpylog) ([#14656](https://github.com/paperless-ngx/paperless-ngx/pull/14656))
+- Fix: ensure import bulk\_edit tasks in Celery workers [@shamoon](https://github.com/shamoon) ([#14653](https://github.com/paperless-ngx/paperless-ngx/pull/14653))
+- Chore(deps): Bump the frontend-angular-dependencies group across 1 directory with 13 updates @[dependabot[bot]](https://github.com/apps/dependabot) ([#14637](https://github.com/paperless-ngx/paperless-ngx/pull/14637))
+- Chore(deps-dev): Bump @playwright/test from 1.62.1 to 1.64.1 in /src-ui @[dependabot[bot]](https://github.com/apps/dependabot) ([#14641](https://github.com/paperless-ngx/paperless-ngx/pull/14641))
+- Chore(deps): Bump pdfjs-dist from 6.3.289 to 6.4.299 in /src-ui @[dependabot[bot]](https://github.com/apps/dependabot) ([#14644](https://github.com/paperless-ngx/paperless-ngx/pull/14644))
+- Chore(deps): Bump uuid from 14.0.2 to 14.0.3 in /src-ui @[dependabot[bot]](https://github.com/apps/dependabot) ([#14640](https://github.com/paperless-ngx/paperless-ngx/pull/14640))
+- Chore(deps-dev): Bump the frontend-eslint-dependencies group across 1 directory with 4 updates @[dependabot[bot]](https://github.com/apps/dependabot) ([#14638](https://github.com/paperless-ngx/paperless-ngx/pull/14638))
+- Chore(deps): Bump marked from 18.0.14 to 18.1.1 in /src-ui @[dependabot[bot]](https://github.com/apps/dependabot) ([#14643](https://github.com/paperless-ngx/paperless-ngx/pull/14643))
+- Chore(deps-dev): Bump prettier from 3.9.9 to 3.9.10 in /src-ui @[dependabot[bot]](https://github.com/apps/dependabot) ([#14645](https://github.com/paperless-ngx/paperless-ngx/pull/14645))
+- Chore(deps-dev): Bump @types/node from 26.6.2 to 26.6.5 in /src-ui @[dependabot[bot]](https://github.com/apps/dependabot) ([#14639](https://github.com/paperless-ngx/paperless-ngx/pull/14639))
+- Fixhancement: add tag names to document history list [@shamoon](https://github.com/shamoon) ([#14625](https://github.com/paperless-ngx/paperless-ngx/pull/14625))
+- Fix: add button type to zoom buttons to prevent form submission [@shamoon](https://github.com/shamoon) ([#14624](https://github.com/paperless-ngx/paperless-ngx/pull/14624))
+- Fix: move document list preview button to own column [@shamoon](https://github.com/shamoon) ([#14610](https://github.com/paperless-ngx/paperless-ngx/pull/14610))
+- Fix: Various small bugs found during typing updates [@stumpylog](https://github.com/stumpylog) ([#14593](https://github.com/paperless-ngx/paperless-ngx/pull/14593))
+- Fix: fix custom fields dropdown text size [@shamoon](https://github.com/shamoon) ([#14606](https://github.com/paperless-ngx/paperless-ngx/pull/14606))
+
+</details>
+
 ## paperless-ngx 3.4.1
 
 ### Bug Fixes

src/documents/consumer.py

@@ -15,7 +15,6 @@ from django.db import transaction
 from django.db.models import Max
 from django.db.models import Q
 from django.utils import timezone
-from filelock import FileLock
 from rest_framework.reverse import reverse
 
 from documents.barcodes import read_barcode_values
@@ -27,7 +26,10 @@ from documents.file_handling import create_source_path_directory
 from documents.file_handling import generate_filename
 from documents.file_handling import generate_unique_filename
 from documents.file_handling import validate_path_in_root
+from documents.file_moves import defer_file_moves
+from documents.file_moves import flush_pending_file_moves
 from documents.loggers import LoggingMixin
+from documents.media_lock import media_lock
 from documents.models import Correspondent
 from documents.models import CustomField
 from documents.models import CustomFieldInstance
@@ -605,7 +607,13 @@ class ConsumerPlugin(
                 # now that everything is done, we can start to store the document
                 # in the system. This will be a transaction and reasonably fast.
                 try:
-                    with transaction.atomic():
+                    # File moves requested in here run after commit and after
+                    # media.lock is released. media.lock is taken before the
+                    # transaction: the version path locks the root row with
+                    # select_for_update, and workflow email actions and file
+                    # placement below re-enter media.lock instead of waiting
+                    # on it while rows are locked.
+                    with defer_file_moves(), media_lock(), transaction.atomic():
                         # store the document.
                         if self.input_doc.root_document_id:
                             # If this is a new version of an existing document, we need
@@ -695,7 +703,7 @@ class ConsumerPlugin(
 
                         # After everything is in the database, copy the files into
                         # place. If this fails, we'll also rollback the transaction.
-                        with FileLock(settings.MEDIA_LOCK):
+                        with media_lock():
                             generated_filename = generate_unique_filename(document)
                             if (
                                 len(str(generated_filename))
@@ -759,10 +767,9 @@ class ConsumerPlugin(
                                     document.archive_path,
                                 )
 
-                        # Don't save with the lock active. Saving will cause the file
-                        # renaming logic to acquire the lock as well.
-                        # This triggers things like file renaming
-                        document.save()
+                        # Any filename follow-up from this save is deferred by
+                        # the scope above until after commit.
+                        document.save(write_file_paths=True)
 
                         if document.root_document_id:
                             document_updated.send(
@@ -803,6 +810,10 @@ class ConsumerPlugin(
                         exception=e,
                     )
 
+        # The storage scope above nests under the consume task's scope, so its
+        # moves are still pending; place the file before the script reads its path.
+        flush_pending_file_moves()
+
         self.run_post_consume_script(document)
 
         self.log.info(f"Document {document} consumption finished")

src/documents/file_moves.py

@@ -0,0 +1,462 @@
+"""
+Document file moves, coalesced to once per unit of work.
+
+Every save that can change a document's generated filename asks for a file
+move through request_file_move(). Inside a scope opened by defer_file_moves()
+(a request, a Celery task, the consumer, or a per-item command step) the
+request is only recorded, and the outermost scope moves each recorded document
+once, after its own writes have committed. Outside any scope the move happens
+immediately.
+
+The mover always takes media.lock first and then reads the document fresh, so
+it starts from the latest committed state and never waits on media.lock while
+its own transaction holds document rows.
+"""
+
+from __future__ import annotations
+
+import logging
+import shutil
+import threading
+from contextlib import asynccontextmanager
+from contextlib import contextmanager
+from contextvars import ContextVar
+from pathlib import Path
+from typing import TYPE_CHECKING
+from typing import Final
+
+from asgiref.sync import sync_to_async
+from django.conf import settings
+from django.db import DatabaseError
+from django.db import transaction
+from django.utils import timezone
+from filelock import Timeout
+
+from documents.caching import clear_document_caches
+from documents.file_handling import create_source_path_directory
+from documents.file_handling import delete_empty_directories
+from documents.file_handling import generate_filename
+from documents.file_handling import generate_unique_filename
+from documents.media_lock import media_lock
+from documents.models import Document
+from documents.utils import compute_checksum
+
+if TYPE_CHECKING:
+    from collections.abc import AsyncIterator
+    from collections.abc import Iterator
+
+logger = logging.getLogger("paperless.handlers")
+
+MEDIA_LOCK_TIMEOUT_SECONDS: Final[int] = 300
+
+
+class CannotMoveFilesException(Exception):
+    pass
+
+
+def in_user_transaction() -> bool:
+    """
+    True when the default connection is inside an atomic block that was not
+    created by Django's test case machinery.
+
+    Django marks the atomic blocks TestCase (and pytest-django) wrap around
+    each test with _from_testcase. Ignoring them keeps tests on the same path
+    as production.
+    """
+    connection = transaction.get_connection()
+    return any(not block._from_testcase for block in connection.atomic_blocks)
+
+
+class FileMoveCollector:
+    """Pending document pks for one file move scope."""
+
+    def __init__(self) -> None:
+        self._pending: set[int] = set()
+        self._lock = threading.Lock()
+        self.closed = False
+
+    def add(self, pk: int) -> bool:
+        """Record a pending move. Returns False if the collector is closed."""
+        with self._lock:
+            if self.closed:
+                return False
+            self._pending.add(pk)
+            return True
+
+    def take_pending(self) -> list[int]:
+        """Remove and return the pending pks, sorted, leaving the collector open."""
+        with self._lock:
+            pending = sorted(self._pending)
+            self._pending.clear()
+            return pending
+
+    def close(self) -> list[int]:
+        """Close the collector and return what was still pending, sorted."""
+        with self._lock:
+            self.closed = True
+            pending = sorted(self._pending)
+            self._pending.clear()
+            return pending
+
+
+# Only the scope context managers call set() / reset(). Everything else reads
+# it and mutates the collector, so adds made in a copied context (thread or
+# task hops) are visible to the scope that flushes.
+_active_collector: ContextVar[FileMoveCollector | None] = ContextVar(
+    "paperless_file_move_collector",
+    default=None,
+)
+
+
+def _path_matches_checksum(path: Path, checksum: str | None) -> bool:
+    if checksum is None or not path.is_file():
+        return False
+
+    return compute_checksum(path) == checksum
+
+
+def move_files_for_document(pk: int) -> Document | None:
+    """
+    Move the document's original and archive to the paths its current
+    committed state generates, then do the same for its versions.
+
+    Returns the fresh instance with its final filename columns, or None when
+    the document does not exist, has no filename yet, or the lock timed out.
+    """
+    # Cheap check without the lock: most saves during consumption happen
+    # before the file is placed. filename is never cleared once set, so this
+    # stays true under the lock. values_list() keeps the check off
+    # Document.global_objects.filter(), which tests patch to inject database
+    # errors into and observe the mover's own writes.
+    filename = (
+        Document.global_objects.values_list("filename", flat=True).filter(pk=pk).first()
+    )
+    if not filename:
+        return None
+
+    try:
+        with media_lock(timeout=MEDIA_LOCK_TIMEOUT_SECONDS):
+            try:
+                document = Document.global_objects.get(pk=pk)
+            except Document.DoesNotExist:
+                logger.debug(f"Document {pk} no longer exists, skipping file move")
+                return None
+            _move_document_files(document)
+    except Timeout:
+        logger.error(
+            f"Timed out after {MEDIA_LOCK_TIMEOUT_SECONDS}s waiting for "
+            f"{settings.MEDIA_LOCK} to move files of document {pk}; the file "
+            f"keeps its current name and will be renamed on its next save",
+        )
+        return None
+
+    # Keep version files in sync with root
+    if document.root_document_id is None:
+        for version_pk in Document.objects.filter(root_document_id=pk).values_list(
+            "pk",
+            flat=True,
+        ):
+            move_files_for_document(version_pk)
+
+    return document
+
+
+def _flush(pks: list[int]) -> None:
+    """Move each document; one failure never stops the rest or raises."""
+    for pk in pks:
+        try:
+            move_files_for_document(pk)
+        except Exception:
+            logger.exception(f"Unexpected error moving files of document {pk}")
+
+
+def _flush_after_commit(pks: list[int]) -> None:
+    if not pks:
+        return
+    if in_user_transaction():
+        transaction.on_commit(lambda: _flush(pks))
+    else:
+        _flush(pks)
+
+
+def _close_and_flush(collector: FileMoveCollector) -> None:
+    _flush_after_commit(collector.close())
+
+
+@contextmanager
+def defer_file_moves() -> Iterator[None]:
+    """
+    Collect file move requests and perform each once when the outermost scope
+    exits, after its own writes have committed.
+    """
+    current = _active_collector.get()
+    if current is not None and not current.closed:
+        yield
+        return
+
+    collector = FileMoveCollector()
+    token = _active_collector.set(collector)
+    try:
+        yield
+    finally:
+        _active_collector.reset(token)
+        _close_and_flush(collector)
+
+
+def flush_pending_file_moves() -> None:
+    """
+    Perform the current scope's pending moves now, keeping the scope open.
+
+    From an eager task inside a request this flushes the request's collector
+    early, which is harmless: moves still run after commit, on fresh state.
+    """
+    collector = _active_collector.get()
+    if collector is None:
+        return
+    _flush_after_commit(collector.take_pending())
+
+
+def _validate_move(
+    document: Document,
+    old_path: Path,
+    new_path: Path,
+    root: Path,
+) -> None:
+    if not new_path.is_relative_to(root):
+        msg = (
+            f"Document {document!s}: Refusing to move file outside root {root}: "
+            f"{new_path}."
+        )
+        logger.warning(msg)
+        raise CannotMoveFilesException(msg)
+
+    if not old_path.is_file():
+        # Can't do anything if the old file does not exist anymore.
+        msg = f"Document {document!s}: File {old_path} doesn't exist."
+        logger.fatal(msg)
+        raise CannotMoveFilesException(msg)
+
+    if new_path.is_file():
+        # Can't do anything if the new file already exists. Skip updating file.
+        msg = f"Document {document!s}: Cannot rename file since target path {new_path} already exists."
+        logger.warning(msg)
+        raise CannotMoveFilesException(msg)
+
+
+def _move_document_files(document: Document) -> None:
+    try:
+        old_filename = document.filename
+        old_source_path = document.source_path
+        move_original = False
+        original_already_moved = False
+
+        old_archive_filename = document.archive_filename
+        old_archive_path = document.archive_path
+        move_archive = False
+        archive_already_moved = False
+
+        candidate_filename = generate_filename(document)
+        if len(str(candidate_filename)) > Document.MAX_STORED_FILENAME_LENGTH:
+            msg = (
+                f"Document {document!s}: Generated filename exceeds db path "
+                f"limit ({len(str(candidate_filename))} > "
+                f"{Document.MAX_STORED_FILENAME_LENGTH}): {candidate_filename!s}"
+            )
+            logger.warning(msg)
+            raise CannotMoveFilesException(msg)
+
+        candidate_source_path = (settings.ORIGINALS_DIR / candidate_filename).resolve()
+        if candidate_filename == Path(old_filename):
+            new_filename = Path(old_filename)
+        elif (
+            candidate_source_path.exists() and candidate_source_path != old_source_path
+        ):
+            if not old_source_path.is_file() and _path_matches_checksum(
+                candidate_source_path,
+                document.checksum,
+            ):
+                new_filename = candidate_filename
+                original_already_moved = True
+            else:
+                # Only fall back to unique search when there is an actual conflict
+                new_filename = generate_unique_filename(document)
+        else:
+            new_filename = candidate_filename
+
+        # Need to convert to string to be able to save it to the db
+        document.filename = str(new_filename)
+        move_original = old_filename != document.filename and not original_already_moved
+
+        if document.has_archive_version:
+            archive_candidate = generate_filename(document, archive_filename=True)
+            if len(str(archive_candidate)) > Document.MAX_STORED_FILENAME_LENGTH:
+                msg = (
+                    f"Document {document!s}: Generated archive filename exceeds "
+                    f"db path limit ({len(str(archive_candidate))} > "
+                    f"{Document.MAX_STORED_FILENAME_LENGTH}): {archive_candidate!s}"
+                )
+                logger.warning(msg)
+                raise CannotMoveFilesException(msg)
+            archive_candidate_path = (
+                settings.ARCHIVE_DIR / archive_candidate
+            ).resolve()
+            if archive_candidate == Path(old_archive_filename):
+                new_archive_filename = Path(old_archive_filename)
+            elif (
+                archive_candidate_path.exists()
+                and archive_candidate_path != old_archive_path
+            ):
+                if not old_archive_path.is_file() and _path_matches_checksum(
+                    archive_candidate_path,
+                    document.archive_checksum,
+                ):
+                    new_archive_filename = archive_candidate
+                    archive_already_moved = True
+                else:
+                    new_archive_filename = generate_unique_filename(
+                        document,
+                        archive_filename=True,
+                    )
+            else:
+                new_archive_filename = archive_candidate
+
+            document.archive_filename = str(new_archive_filename)
+
+            move_archive = (
+                old_archive_filename != document.archive_filename
+                and not archive_already_moved
+            )
+        else:
+            move_archive = False
+
+        if not move_original and not move_archive:
+            updates: dict[str, object] = {}
+            if old_filename != document.filename:
+                updates["filename"] = document.filename
+            if old_archive_filename != document.archive_filename:
+                updates["archive_filename"] = document.archive_filename
+            if updates:
+                # global_objects so a trashed document's path is fixed too.
+                updates["modified"] = timezone.now()
+                # Don't save() here to prevent infinite recursion.
+                Document.global_objects.filter(pk=document.pk).update(**updates)
+            return
+
+        if move_original:
+            _validate_move(
+                document,
+                old_source_path,
+                document.source_path,
+                settings.ORIGINALS_DIR,
+            )
+            create_source_path_directory(document.source_path)
+            shutil.move(old_source_path, document.source_path)
+
+        if move_archive:
+            _validate_move(
+                document,
+                old_archive_path,
+                document.archive_path,
+                settings.ARCHIVE_DIR,
+            )
+            create_source_path_directory(document.archive_path)
+            shutil.move(old_archive_path, document.archive_path)
+
+        # Don't save() here to prevent infinite recursion.
+        Document.global_objects.filter(pk=document.pk).update(
+            filename=document.filename,
+            archive_filename=document.archive_filename,
+            modified=timezone.now(),
+        )
+        # Clear any caching for this document.  Slightly overkill, but not terrible
+        clear_document_caches(document.pk)
+
+    except (OSError, DatabaseError, CannotMoveFilesException) as e:
+        logger.warning(f"Exception during file handling: {e}")
+        # This happens when either:
+        #  - moving the files failed due to file system errors
+        #  - saving to the database failed due to database errors
+        # In both cases, we need to revert to the original state.
+
+        # Try to move files to their original location.
+        try:
+            if move_original and document.source_path.is_file():
+                logger.info("Restoring previous original path")
+                shutil.move(document.source_path, old_source_path)
+
+            if move_archive and document.archive_path.is_file():
+                logger.info("Restoring previous archive path")
+                shutil.move(document.archive_path, old_archive_path)
+
+        except Exception:
+            # This is fine, since:
+            # A: if we managed to move source from A to B, we will also
+            #  manage to move it from B to A. If not, we have a serious
+            #  issue that's going to get caught by the santiy checker.
+            #  All files remain in place and will never be overwritten,
+            #  so this is not the end of the world.
+            # B: if moving the original file failed, nothing has changed
+            #  anyway.
+            pass
+
+        # restore old values on the document
+        document.filename = old_filename
+        document.archive_filename = old_archive_filename
+
+    # finally, remove any empty sub folders. This will do nothing if
+    # something has failed above.
+    if not old_source_path.is_file():
+        delete_empty_directories(
+            Path(old_source_path).parent,
+            root=settings.ORIGINALS_DIR,
+        )
+
+    if document.has_archive_version and not old_archive_path.is_file():
+        delete_empty_directories(
+            Path(old_archive_path).parent,
+            root=settings.ARCHIVE_DIR,
+        )
+
+
+def request_file_move(document: Document) -> None:
+    """
+    Ask for the document's files to follow its committed state.
+
+    Inside an open scope the request is recorded and performed when the scope
+    ends. Otherwise the move happens now and only the final filename columns
+    are copied onto the caller's instance, so source_path is current after
+    save() while other in-memory fields are left alone.
+    """
+    collector = _active_collector.get()
+    if collector is not None and collector.add(document.pk):
+        return
+
+    moved = move_files_for_document(document.pk)
+    if moved is not None:
+        document.filename = moved.filename
+        document.archive_filename = moved.archive_filename
+
+
+@asynccontextmanager
+async def adefer_file_moves() -> AsyncIterator[None]:
+    """
+    Async twin of defer_file_moves().
+
+    When something was collected, the flush runs through sync_to_async: its
+    transaction check needs the thread owning the DB connection, and the
+    blocking media.lock must never run on the event loop.
+    """
+    current = _active_collector.get()
+    if current is not None and not current.closed:
+        yield
+        return
+
+    collector = FileMoveCollector()
+    token = _active_collector.set(collector)
+    try:
+        yield
+    finally:
+        _active_collector.reset(token)
+        pks = collector.close()
+        if pks:
+            await sync_to_async(_flush_after_commit, thread_sensitive=True)(pks)

src/documents/mail.py

@@ -4,9 +4,9 @@ from dataclasses import dataclass
 from email import message_from_bytes
 from pathlib import Path
 
-from django.conf import settings
 from django.core.mail import EmailMessage
-from filelock import FileLock
+
+from documents.media_lock import media_lock
 
 
 @dataclass(frozen=True)
@@ -48,7 +48,7 @@ def send_email(
     used_filenames: set[str] = set()
 
     # Something could be renaming the file concurrently so it can't be attached
-    with FileLock(settings.MEDIA_LOCK):
+    with media_lock():
         for attachment in attachments:
             filename = _get_unique_filename(
                 attachment.friendly_name,

src/documents/management/commands/document_exporter.py

@@ -18,10 +18,11 @@ from django.core import serializers
 from django.core.management.base import CommandError
 from django.db import transaction
 from django.utils import timezone
-from filelock import FileLock
 from guardian.models import GroupObjectPermission
 from guardian.models import UserObjectPermission
 
+from documents.media_lock import media_lock
+
 if TYPE_CHECKING:
     from collections.abc import Generator
 
@@ -328,7 +329,7 @@ class Command(CryptMixin, PaperlessCommand):
             )
 
         # Prevent any ongoing changes in the documents while exporting
-        with FileLock(settings.MEDIA_LOCK), sink:
+        with media_lock(), sink:
             self.dump(sink)
 
     def dump(self, sink: ExportSink) -> None:

src/documents/management/commands/document_importer.py

@@ -29,7 +29,6 @@ from django.db.models import GeneratedField
 from django.db.models import Model
 from django.db.models.signals import m2m_changed
 from django.db.models.signals import post_save
-from filelock import FileLock
 from guardian.shortcuts import clear_ct_cache
 
 from documents.export.compression import compress_type_readable
@@ -37,6 +36,7 @@ from documents.export.compression import unreadable_method_names
 from documents.file_handling import create_source_path_directory
 from documents.management.commands.base import PaperlessCommand
 from documents.management.commands.mixins import CryptMixin
+from documents.media_lock import media_lock
 from documents.models import Correspondent
 from documents.models import CustomField
 from documents.models import CustomFieldInstance
@@ -644,7 +644,7 @@ class Command(CryptMixin, PaperlessCommand):
             else:
                 archive_path = None
 
-            with FileLock(settings.MEDIA_LOCK):
+            with media_lock():
                 if Path(document.source_path).is_file():
                     raise FileExistsError(document.source_path)
 
@@ -678,7 +678,7 @@ class Command(CryptMixin, PaperlessCommand):
                     f"Share link bundle {bundle.pk} does not have a valid file path.",
                 )
 
-            with FileLock(settings.MEDIA_LOCK):
+            with media_lock():
                 bundle_target_path.parent.mkdir(parents=True, exist_ok=True)
                 copy_file_with_basic_stats(
                     bundle_source_path,

src/documents/management/commands/document_renamer.py

@@ -1,5 +1,4 @@
-from django.db.models.signals import post_save
-
+from documents.file_moves import move_files_for_document
 from documents.management.commands.base import PaperlessCommand
 from documents.models import Document
 
@@ -11,5 +10,8 @@ class Command(PaperlessCommand):
     supports_multiprocessing = False
 
     def handle(self, *args, **options):
-        for document in self.track(Document.objects.all(), description="Renaming..."):
-            post_save.send(Document, instance=document, created=False)
+        for document in self.track(
+            Document.objects.only("pk"),
+            description="Renaming...",
+        ):
+            move_files_for_document(document.pk)

src/documents/management/commands/document_retagger.py

@@ -9,6 +9,7 @@ from rich.table import Table
 from rich.text import Text
 
 from documents.classifier import load_classifier
+from documents.file_moves import defer_file_moves
 from documents.management.commands.base import PaperlessCommand
 from documents.models import Document
 from documents.signals.handlers import set_correspondent
@@ -273,61 +274,63 @@ class Command(PaperlessCommand):
                 description="Retagging...",
                 stats_renderer=render_stats,
             ):
-                suggestion = DocumentSuggestion(document=document)
+                # One file move per document, after all of its assignments.
+                with defer_file_moves():
+                    suggestion = DocumentSuggestion(document=document)
 
-                if do_correspondent:
-                    correspondent = set_correspondent(
-                        None,
-                        document,
-                        classifier=classifier,
-                        replace=overwrite,
-                        use_first=use_first,
-                        dry_run=suggest,
-                    )
-                    if correspondent is not None:
-                        stats.correspondents += 1
-                        suggestion.correspondent = correspondent
+                    if do_correspondent:
+                        correspondent = set_correspondent(
+                            None,
+                            document,
+                            classifier=classifier,
+                            replace=overwrite,
+                            use_first=use_first,
+                            dry_run=suggest,
+                        )
+                        if correspondent is not None:
+                            stats.correspondents += 1
+                            suggestion.correspondent = correspondent
 
-                if do_document_type:
-                    document_type = set_document_type(
-                        None,
-                        document,
-                        classifier=classifier,
-                        replace=overwrite,
-                        use_first=use_first,
-                        dry_run=suggest,
-                    )
-                    if document_type is not None:
-                        stats.document_types += 1
-                        suggestion.document_type = document_type
+                    if do_document_type:
+                        document_type = set_document_type(
+                            None,
+                            document,
+                            classifier=classifier,
+                            replace=overwrite,
+                            use_first=use_first,
+                            dry_run=suggest,
+                        )
+                        if document_type is not None:
+                            stats.document_types += 1
+                            suggestion.document_type = document_type
 
-                if do_tags:
-                    tags_to_add, tags_to_remove = set_tags(
-                        None,
-                        document,
-                        classifier=classifier,
-                        replace=overwrite,
-                        dry_run=suggest,
-                    )
-                    stats.tags_added += len(tags_to_add)
-                    stats.tags_removed += len(tags_to_remove)
-                    suggestion.tags_to_add = frozenset(tags_to_add)
-                    suggestion.tags_to_remove = frozenset(tags_to_remove)
+                    if do_tags:
+                        tags_to_add, tags_to_remove = set_tags(
+                            None,
+                            document,
+                            classifier=classifier,
+                            replace=overwrite,
+                            dry_run=suggest,
+                        )
+                        stats.tags_added += len(tags_to_add)
+                        stats.tags_removed += len(tags_to_remove)
+                        suggestion.tags_to_add = frozenset(tags_to_add)
+                        suggestion.tags_to_remove = frozenset(tags_to_remove)
 
-                if do_storage_path:
-                    storage_path = set_storage_path(
-                        None,
-                        document,
-                        classifier=classifier,
-                        replace=overwrite,
-                        use_first=use_first,
-                        dry_run=suggest,
-                    )
-                    if storage_path is not None:
-                        stats.storage_paths += 1
-                        suggestion.storage_path = storage_path
+                    if do_storage_path:
+                        storage_path = set_storage_path(
+                            None,
+                            document,
+                            classifier=classifier,
+                            replace=overwrite,
+                            use_first=use_first,
+                            dry_run=suggest,
+                        )
+                        if storage_path is not None:
+                            stats.storage_paths += 1
+                            suggestion.storage_path = storage_path
 
-                stats.documents_processed += 1
+                    stats.documents_processed += 1
 
                 if suggest:
                     suggestions.append(suggestion)

src/documents/media_lock.py

@@ -0,0 +1,36 @@
+"""
+The global media lock, reentrant within a thread.
+
+Code that both needs media.lock and opens a transaction that locks document
+rows must take media.lock first, so no transaction ever waits on it while
+holding rows. Reentrancy lets code already holding the lock (for example the
+consumer) call into code that takes it again (for example a workflow email
+action attaching files) without blocking itself.
+"""
+
+from __future__ import annotations
+
+from contextlib import contextmanager
+from typing import TYPE_CHECKING
+
+from django.conf import settings
+from filelock import FileLock
+
+if TYPE_CHECKING:
+    from collections.abc import Iterator
+
+
+@contextmanager
+def media_lock(timeout: float = -1) -> Iterator[None]:
+    """
+    Hold settings.MEDIA_LOCK for the block.
+
+    is_singleton returns one FileLock instance per path in this process, and
+    the default thread_local=True keeps the hold count per thread, so the
+    same thread re-enters while other threads and processes are excluded.
+    The timeout goes to acquire() because a singleton rejects construction
+    with different parameters. A negative timeout waits forever.
+    """
+    lock = FileLock(settings.MEDIA_LOCK, is_singleton=True)
+    with lock.acquire(timeout=timeout):
+        yield

src/documents/models.py

@@ -1,6 +1,7 @@
 import datetime
 import uuid
 from pathlib import Path
+from typing import Any
 from typing import Final
 
 import pathvalidate
@@ -157,6 +158,12 @@ class StoragePath(MatchingModel):
 
 class Document(SoftDeleteModel, ModelWithOwner):  # type: ignore[django-manager-missing]
     MAX_STORED_FILENAME_LENGTH: Final[int] = 1024
+    # Written only by the file mover, the consumer's file placement and the
+    # archiver. A full save() of a stale instance must never write them, or
+    # the DB can point at a path the file has already been moved away from.
+    FILE_PATH_FIELDS: Final[frozenset[str]] = frozenset(
+        {"filename", "archive_filename"},
+    )
     # Per-component limit of common filesystems (ext4, XFS, btrfs, ZFS, APFS)
     MAX_PUBLIC_FILENAME_BYTES: Final[int] = 255
 
@@ -368,6 +375,31 @@ class Document(SoftDeleteModel, ModelWithOwner):  # type: ignore[django-manager-
             res += f" {self.title}"
         return res
 
+    def save(self, *, write_file_paths: bool = False, **kwargs: Any) -> None:
+        """
+        Save the document without writing filename / archive_filename on an
+        existing row, unless write_file_paths=True or update_fields names them.
+
+        This mirrors how Django narrows a save on a deferred instance: only the
+        loaded, concrete, non-generated, non-primary-key fields are written.
+        """
+        if (
+            not write_file_paths
+            and not self._state.adding
+            and kwargs.get("update_fields") is None
+            and not kwargs.get("force_insert", False)
+        ):
+            deferred = self.get_deferred_fields()
+            kwargs["update_fields"] = [
+                field.attname
+                for field in self._meta.concrete_fields
+                if not field.primary_key
+                and not field.generated
+                and field.name not in self.FILE_PATH_FIELDS
+                and field.attname not in deferred
+            ]
+        super().save(**kwargs)
+
     def get_effective_barcodes(self) -> list["DocumentBarcode"]:
         """
         Returns the stored barcodes for the document, like

src/documents/signals/handlers.py

@@ -19,25 +19,22 @@ from celery.signals import worker_process_shutdown
 from django.conf import settings
 from django.contrib.auth.models import Group
 from django.contrib.auth.models import User
-from django.db import DatabaseError
 from django.db import close_old_connections
 from django.db import connections
 from django.db import models
 from django.db.models import Q
 from django.dispatch import receiver
 from django.utils import timezone
-from filelock import FileLock
 from rest_framework import serializers
 
 from documents import matching
-from documents.caching import clear_document_caches
 from documents.caching import invalidate_llm_suggestions_cache
 from documents.caching import invalidate_suggestions_cache
 from documents.data_models import ConsumableDocument
-from documents.file_handling import create_source_path_directory
 from documents.file_handling import delete_empty_directories
-from documents.file_handling import generate_filename
-from documents.file_handling import generate_unique_filename
+from documents.file_moves import move_files_for_document
+from documents.file_moves import request_file_move
+from documents.media_lock import media_lock
 from documents.models import Correspondent
 from documents.models import CustomField
 from documents.models import CustomFieldInstance
@@ -55,7 +52,6 @@ from documents.models import WorkflowTrigger
 from documents.permissions import restrict_queryset_to_visible
 from documents.plugins.helpers import DocumentsStatusManager
 from documents.templating.utils import convert_format_str_to_template_format
-from documents.utils import compute_checksum
 from documents.utils import copy_file_with_basic_stats
 from documents.workflows.actions import build_workflow_action_context
 from documents.workflows.actions import execute_email_action
@@ -339,7 +335,7 @@ def set_storage_path(
 
 # see empty_trash in documents/tasks.py for signal handling
 def cleanup_document_deletion(sender, instance, **kwargs) -> None:
-    with FileLock(settings.MEDIA_LOCK):
+    with media_lock():
         if settings.EMPTY_TRASH_DIR:
             # Find a non-conflicting filename in case a document with the same
             # name was moved to trash earlier
@@ -405,17 +401,6 @@ def cleanup_document_deletion(sender, instance, **kwargs) -> None:
             )
 
 
-class CannotMoveFilesException(Exception):
-    pass
-
-
-def _path_matches_checksum(path: Path, checksum: str | None) -> bool:
-    if checksum is None or not path.is_file():
-        return False
-
-    return compute_checksum(path) == checksum
-
-
 def _filename_template_uses_custom_fields(doc: Document) -> bool:
     template = None
     if doc.storage_path is not None:
@@ -438,235 +423,21 @@ def update_filename_and_move_files(
     instance: Document | CustomFieldInstance,
     **kwargs,
 ) -> None:
+    if kwargs.get("reverse"):
+        # tag.documents.add(doc): instance is the Tag, not a Document.
+        return
+    action = kwargs.get("action")
+    if action is not None and action not in {"post_add", "post_remove", "post_clear"}:
+        # m2m_changed fires pre_* and post_* actions; only post_* describe the
+        # relation state the filename should follow.
+        return
+
     if isinstance(instance, CustomFieldInstance):
         if not _filename_template_uses_custom_fields(instance.document):
             return
         instance = instance.document
 
-    def validate_move(instance, old_path: Path, new_path: Path, root: Path) -> None:
-        if not new_path.is_relative_to(root):
-            msg = (
-                f"Document {instance!s}: Refusing to move file outside root {root}: "
-                f"{new_path}."
-            )
-            logger.warning(msg)
-            raise CannotMoveFilesException(msg)
-
-        if not old_path.is_file():
-            # Can't do anything if the old file does not exist anymore.
-            msg = f"Document {instance!s}: File {old_path} doesn't exist."
-            logger.fatal(msg)
-            raise CannotMoveFilesException(msg)
-
-        if new_path.is_file():
-            # Can't do anything if the new file already exists. Skip updating file.
-            msg = f"Document {instance!s}: Cannot rename file since target path {new_path} already exists."
-            logger.warning(msg)
-            raise CannotMoveFilesException(msg)
-
-    if not instance.filename:
-        # Can't update the filename if there is no filename to begin with
-        # This happens when the consumer creates a new document.
-        # The document is modified and saved multiple times, and only after
-        # everything is done (i.e., the generated filename is final),
-        # filename will be set to the location where the consumer has put
-        # the file.
-        #
-        # This will in turn cause this logic to move the file where it belongs.
-        return
-
-    with FileLock(settings.MEDIA_LOCK):
-        # If this was waiting for the lock, the filename or archive_filename
-        # of this document may have been updated.  This happens if multiple updates
-        # get queued from the UI for the same document
-        # So freshen up the data before doing anything
-        instance.refresh_from_db()
-
-        try:
-            old_filename = instance.filename
-            old_source_path = instance.source_path
-            move_original = False
-            original_already_moved = False
-
-            old_archive_filename = instance.archive_filename
-            old_archive_path = instance.archive_path
-            move_archive = False
-            archive_already_moved = False
-
-            candidate_filename = generate_filename(instance)
-            if len(str(candidate_filename)) > Document.MAX_STORED_FILENAME_LENGTH:
-                msg = (
-                    f"Document {instance!s}: Generated filename exceeds db path "
-                    f"limit ({len(str(candidate_filename))} > "
-                    f"{Document.MAX_STORED_FILENAME_LENGTH}): {candidate_filename!s}"
-                )
-                logger.warning(msg)
-                raise CannotMoveFilesException(msg)
-
-            candidate_source_path = (
-                settings.ORIGINALS_DIR / candidate_filename
-            ).resolve()
-            if candidate_filename == Path(old_filename):
-                new_filename = Path(old_filename)
-            elif (
-                candidate_source_path.exists()
-                and candidate_source_path != old_source_path
-            ):
-                if not old_source_path.is_file() and _path_matches_checksum(
-                    candidate_source_path,
-                    instance.checksum,
-                ):
-                    new_filename = candidate_filename
-                    original_already_moved = True
-                else:
-                    # Only fall back to unique search when there is an actual conflict
-                    new_filename = generate_unique_filename(instance)
-            else:
-                new_filename = candidate_filename
-
-            # Need to convert to string to be able to save it to the db
-            instance.filename = str(new_filename)
-            move_original = (
-                old_filename != instance.filename and not original_already_moved
-            )
-
-            if instance.has_archive_version:
-                archive_candidate = generate_filename(instance, archive_filename=True)
-                if len(str(archive_candidate)) > Document.MAX_STORED_FILENAME_LENGTH:
-                    msg = (
-                        f"Document {instance!s}: Generated archive filename exceeds "
-                        f"db path limit ({len(str(archive_candidate))} > "
-                        f"{Document.MAX_STORED_FILENAME_LENGTH}): {archive_candidate!s}"
-                    )
-                    logger.warning(msg)
-                    raise CannotMoveFilesException(msg)
-                archive_candidate_path = (
-                    settings.ARCHIVE_DIR / archive_candidate
-                ).resolve()
-                if archive_candidate == Path(old_archive_filename):
-                    new_archive_filename = Path(old_archive_filename)
-                elif (
-                    archive_candidate_path.exists()
-                    and archive_candidate_path != old_archive_path
-                ):
-                    if not old_archive_path.is_file() and _path_matches_checksum(
-                        archive_candidate_path,
-                        instance.archive_checksum,
-                    ):
-                        new_archive_filename = archive_candidate
-                        archive_already_moved = True
-                    else:
-                        new_archive_filename = generate_unique_filename(
-                            instance,
-                            archive_filename=True,
-                        )
-                else:
-                    new_archive_filename = archive_candidate
-
-                instance.archive_filename = str(new_archive_filename)
-
-                move_archive = (
-                    old_archive_filename != instance.archive_filename
-                    and not archive_already_moved
-                )
-            else:
-                move_archive = False
-
-            if not move_original and not move_archive:
-                updates = {"modified": timezone.now()}
-                if old_filename != instance.filename:
-                    updates["filename"] = instance.filename
-                if old_archive_filename != instance.archive_filename:
-                    updates["archive_filename"] = instance.archive_filename
-
-                # Don't save() here to prevent infinite recursion.
-                Document.objects.filter(pk=instance.pk).update(**updates)
-                return
-
-            if move_original:
-                validate_move(
-                    instance,
-                    old_source_path,
-                    instance.source_path,
-                    settings.ORIGINALS_DIR,
-                )
-                create_source_path_directory(instance.source_path)
-                shutil.move(old_source_path, instance.source_path)
-
-            if move_archive:
-                validate_move(
-                    instance,
-                    old_archive_path,
-                    instance.archive_path,
-                    settings.ARCHIVE_DIR,
-                )
-                create_source_path_directory(instance.archive_path)
-                shutil.move(old_archive_path, instance.archive_path)
-
-            # Don't save() here to prevent infinite recursion.
-            Document.global_objects.filter(pk=instance.pk).update(
-                filename=instance.filename,
-                archive_filename=instance.archive_filename,
-                modified=timezone.now(),
-            )
-            # Clear any caching for this document.  Slightly overkill, but not terrible
-            clear_document_caches(instance.pk)
-
-        except (OSError, DatabaseError, CannotMoveFilesException) as e:
-            logger.warning(f"Exception during file handling: {e}")
-            # This happens when either:
-            #  - moving the files failed due to file system errors
-            #  - saving to the database failed due to database errors
-            # In both cases, we need to revert to the original state.
-
-            # Try to move files to their original location.
-            try:
-                if move_original and instance.source_path.is_file():
-                    logger.info("Restoring previous original path")
-                    shutil.move(instance.source_path, old_source_path)
-
-                if move_archive and instance.archive_path.is_file():
-                    logger.info("Restoring previous archive path")
-                    shutil.move(instance.archive_path, old_archive_path)
-
-            except Exception:
-                # This is fine, since:
-                # A: if we managed to move source from A to B, we will also
-                #  manage to move it from B to A. If not, we have a serious
-                #  issue that's going to get caught by the santiy checker.
-                #  All files remain in place and will never be overwritten,
-                #  so this is not the end of the world.
-                # B: if moving the original file failed, nothing has changed
-                #  anyway.
-                pass
-
-            # restore old values on the instance
-            instance.filename = old_filename
-            instance.archive_filename = old_archive_filename
-
-        # finally, remove any empty sub folders. This will do nothing if
-        # something has failed above.
-        if not old_source_path.is_file():
-            delete_empty_directories(
-                Path(old_source_path).parent,
-                root=settings.ORIGINALS_DIR,
-            )
-
-        if instance.has_archive_version and not old_archive_path.is_file():
-            delete_empty_directories(
-                Path(old_archive_path).parent,
-                root=settings.ARCHIVE_DIR,
-            )
-
-    # Keep version files in sync with root
-    if instance.root_document_id is None:
-        for version_doc in Document.objects.filter(root_document_id=instance.pk).only(
-            "pk",
-        ):
-            update_filename_and_move_files(
-                Document,
-                version_doc,
-            )
+    request_file_move(instance)
 
 
 @shared_task
@@ -691,7 +462,8 @@ def process_cf_select_update(custom_field: CustomField) -> None:
 
     for cf_instance in custom_field.fields.select_related("document").iterator():
         # Update the filename and move files if necessary
-        update_filename_and_move_files(CustomFieldInstance, cf_instance)
+        if _filename_template_uses_custom_fields(cf_instance.document):
+            move_files_for_document(cf_instance.document_id)
 
 
 # should be disabled in /src/documents/management/commands/document_importer.py handle

src/documents/tasks.py

@@ -16,7 +16,6 @@ from django.db import models
 from django.db import transaction
 from django.db.models.signals import post_save
 from django.utils import timezone
-from filelock import FileLock
 
 from documents import sanity_checker
 from documents.barcodes import BarcodePlugin
@@ -42,6 +41,7 @@ from documents.double_sided import CollatePlugin
 from documents.file_handling import create_source_path_directory
 from documents.file_handling import generate_unique_filename
 from documents.matching import prefilter_documents_by_workflowtrigger
+from documents.media_lock import media_lock
 from documents.models import Correspondent
 from documents.models import CustomFieldInstance
 from documents.models import Document
@@ -310,6 +310,7 @@ def sanity_check(*, raise_on_error: bool = True) -> str:
 
 @shared_task
 def bulk_update_documents(document_ids) -> None:
+    from documents.file_moves import flush_pending_file_moves
     from documents.search import get_backend
 
     document_ids = list(document_ids)
@@ -330,6 +331,9 @@ def bulk_update_documents(document_ids) -> None:
             skip_ai_index=True,  # bulk path calls update_llm_index once below
         )
         post_save.send(Document, instance=doc, created=False)
+        # Move this document's files now rather than at the end of the task,
+        # so a long bulk edit makes steady progress.
+        flush_pending_file_moves()
 
     with get_backend().batch_update() as batch:
         batch.add_or_update_ids(document_ids)
@@ -417,7 +421,13 @@ def update_document_content_maybe_archive_file(
 
             thumbnail = parser.get_thumbnail(document.source_path, mime_type)
 
-            with transaction.atomic():
+            # media.lock first, then the transaction: never wait on media.lock
+            # while this transaction holds the document row. Also keeps the
+            # unique archive filename generation under the lock.
+            with media_lock(), transaction.atomic():
+                # Loaded before the (long) parse; read current metadata now
+                # that the lock is held, for the archive filename.
+                document.refresh_from_db()
                 oldDocument = Document.objects.get(pk=document.pk)
                 if parser.get_archive_path():
                     checksum = compute_checksum(parser.get_archive_path())
@@ -482,11 +492,10 @@ def update_document_content_maybe_archive_file(
                         modified=timezone.now(),
                     )
 
-                with FileLock(settings.MEDIA_LOCK):
-                    if parser.get_archive_path():
-                        create_source_path_directory(document.archive_path)
-                        shutil.move(parser.get_archive_path(), document.archive_path)
-                    shutil.move(thumbnail, document.thumbnail_path)
+                if parser.get_archive_path():
+                    create_source_path_directory(document.archive_path)
+                    shutil.move(parser.get_archive_path(), document.archive_path)
+                shutil.move(thumbnail, document.thumbnail_path)
 
             document.refresh_from_db()
             logger.info(
@@ -531,7 +540,11 @@ def empty_trash(doc_ids=None) -> None:
         deleted_document_ids = list(documents.values_list("id", flat=True))
         # Temporarily connect the cleanup handler
         models.signals.post_delete.connect(cleanup_document_deletion, sender=Document)
-        documents.delete()  # this is effectively a hard delete
+        # media.lock before Django's delete transaction: post_delete runs
+        # cleanup_document_deletion inside it while the rows are locked, and
+        # that handler re-enters the lock instead of waiting.
+        with media_lock():
+            documents.delete()  # this is effectively a hard delete
         logger.info(f"Deleted {len(deleted_document_ids)} documents from trash")
 
         if settings.AUDIT_LOG_ENABLED:
@@ -561,6 +574,8 @@ def check_scheduled_workflows() -> None:
 
     Once a document satisfies this condition, and recurring/non-recurring constraints are met, the workflow is run.
     """
+    from documents.file_moves import flush_pending_file_moves
+
     scheduled_workflows = get_workflows_for_trigger(
         WorkflowTrigger.WorkflowTriggerType.SCHEDULED,
     )
@@ -688,6 +703,9 @@ def check_scheduled_workflows() -> None:
                             sender=None,
                             document=document,
                         )
+                        # Move this document's files now rather than at the end
+                        # of the task, so a long run makes steady progress.
+                        flush_pending_file_moves()
 
 
 def update_document_parent_tags(tag: Tag, new_parent: Tag) -> None:

src/documents/tests/test_bulk_edit.py

@@ -739,7 +739,7 @@ class TestPDFActions(DirectoriesMixin, TestCase):
             mime_type="application/pdf",
         )
         self.doc1.archive_filename = sample1_archive
-        self.doc1.save()
+        self.doc1.save(write_file_paths=True)
         self.doc2 = Document.objects.create(
             checksum="B",
             title="B",
@@ -749,7 +749,7 @@ class TestPDFActions(DirectoriesMixin, TestCase):
             created=date(2023, 1, 2),
         )
         self.doc2.archive_filename = sample2_archive
-        self.doc2.save()
+        self.doc2.save(write_file_paths=True)
         self.doc3 = Document.objects.create(
             checksum="C",
             title="C",
@@ -774,7 +774,7 @@ class TestPDFActions(DirectoriesMixin, TestCase):
             created=date(2023, 1, 3),
         )
         self.img_doc.archive_filename = img_doc_archive
-        self.img_doc.save()
+        self.img_doc.save(write_file_paths=True)
 
     @staticmethod
     def fake_decrypt(

src/documents/tests/test_consumer.py

@@ -27,6 +27,10 @@ from documents.models import Document
 from documents.models import DocumentType
 from documents.models import StoragePath
 from documents.models import Tag
+from documents.models import Workflow
+from documents.models import WorkflowAction
+from documents.models import WorkflowActionEmail
+from documents.models import WorkflowTrigger
 from documents.parsers import ParseError
 from documents.plugins.helpers import ProgressStatusOptions
 from documents.tasks import sanity_check
@@ -38,6 +42,7 @@ from paperless_testing.factories import DocumentFactory
 from paperless_testing.factories import TagFactory
 from paperless_testing.factories import UserFactory
 from paperless_testing.fakes.progress import FakeProgressManager
+from paperless_testing.locks import record_media_lock_acquisitions
 from paperless_testing.permissions import grant_object
 from paperless_testing.samples import MULTI_PAGE_DIGITAL_PDF
 from paperless_testing.samples import MULTI_PAGE_IMAGES_PDF
@@ -221,6 +226,160 @@ class TestConsumer(
         shutil.copy(src, dst)
         return dst
 
+    @mock.patch("documents.consumer.load_classifier")
+    def test_consume_takes_media_lock_outside_transaction(self, m) -> None:
+        """
+        GIVEN:
+            - A new document to consume
+        WHEN:
+            - It is consumed
+        THEN:
+            - media.lock is acquired
+            - media.lock is never first acquired inside a transaction
+            - The document's file is in place
+        """
+        m.return_value = MagicMock()
+
+        with record_media_lock_acquisitions() as acquisitions:
+            with self.get_consumer(self.get_test_file()) as consumer:
+                consumer.run()
+
+        self.assertTrue(acquisitions)
+        self.assertFalse(any(a.in_user_transaction for a in acquisitions))
+        document = Document.objects.first()
+        assert document is not None
+        self.assertIsFile(document.source_path)
+
+    @mock.patch("documents.consumer.load_classifier")
+    def test_consume_places_file_before_post_consume_script(self, m) -> None:
+        """
+        GIVEN:
+            - A new document to consume
+        WHEN:
+            - It is consumed
+        THEN:
+            - Pending file moves are flushed before the post-consume script runs
+        """
+        m.return_value = MagicMock()
+        order = MagicMock()
+
+        with (
+            mock.patch(
+                "documents.consumer.flush_pending_file_moves",
+                order.flush,
+            ),
+            mock.patch.object(
+                ConsumerPlugin,
+                "run_post_consume_script",
+                order.script,
+            ),
+        ):
+            with self.get_consumer(self.get_test_file()) as consumer:
+                consumer.run()
+
+        self.assertEqual(
+            [c[0] for c in order.mock_calls],
+            ["flush", "script"],
+        )
+
+    @mock.patch("documents.consumer.load_classifier")
+    def test_consume_version_takes_media_lock_outside_transaction(self, m) -> None:
+        """
+        GIVEN:
+            - An existing root document
+        WHEN:
+            - A new version of it is consumed (the root row is locked with
+              select_for_update inside the consumer's transaction)
+        THEN:
+            - media.lock is never first acquired inside a transaction
+            - The root's modified is bumped
+        """
+        m.return_value = MagicMock()
+
+        with self.get_consumer(self.get_test_file()) as consumer:
+            consumer.run()
+        root_doc = Document.objects.first()
+        assert root_doc is not None
+        original_modified = timezone.now() - datetime.timedelta(days=1)
+        Document.objects.filter(pk=root_doc.pk).update(modified=original_modified)
+
+        version_file = self.get_test_file2()
+        status = FakeProgressManager(version_file.name, None)
+        overrides = DocumentMetadataOverrides(version_label="v2")
+        doc = ConsumableDocument(
+            DocumentSource.ApiUpload,
+            original_file=version_file,
+            root_document_id=root_doc.pk,
+        )
+        preflight = ConsumerPreflightPlugin(
+            doc,
+            overrides,
+            status,  # type: ignore[arg-type]
+            self.dirs.scratch_dir,
+            "task-id",
+        )
+        preflight.setup()
+        preflight.run()
+        consumer = ConsumerPlugin(
+            doc,
+            overrides,
+            status,  # type: ignore[arg-type]
+            self.dirs.scratch_dir,
+            "task-id",
+        )
+        consumer.setup()
+
+        try:
+            with record_media_lock_acquisitions() as acquisitions:
+                consumer.run()
+        finally:
+            consumer.cleanup()
+
+        self.assertTrue(acquisitions)
+        self.assertFalse(any(a.in_user_transaction for a in acquisitions))
+        root_doc.refresh_from_db()
+        self.assertGreater(root_doc.modified, original_modified)
+
+    @override_settings(
+        PAPERLESS_EMAIL_HOST="localhost",
+        EMAIL_ENABLED=True,
+        PAPERLESS_URL="http://localhost:8000",
+    )
+    @mock.patch("documents.consumer.load_classifier")
+    @mock.patch("django.core.mail.message.EmailMessage.send", return_value=1)
+    def test_consumption_workflow_email_runs_under_held_lock(self, send, m) -> None:
+        """
+        GIVEN:
+            - A document added workflow with an email action
+        WHEN:
+            - A document is consumed (the consumer holds media.lock while
+              workflows run)
+        THEN:
+            - The email action re-enters media.lock and sends without blocking
+        """
+        m.return_value = MagicMock()
+        trigger = WorkflowTrigger.objects.create(
+            type=WorkflowTrigger.WorkflowTriggerType.DOCUMENT_ADDED,
+        )
+        email = WorkflowActionEmail.objects.create(
+            subject="added",
+            body="body",
+            to="someone@example.com",
+            include_document=True,
+        )
+        action = WorkflowAction.objects.create(
+            type=WorkflowAction.WorkflowActionType.EMAIL,
+            email=email,
+        )
+        workflow = Workflow.objects.create(name="email on add")
+        workflow.triggers.add(trigger)
+        workflow.actions.add(action)
+
+        with self.get_consumer(self.get_test_file()) as consumer:
+            consumer.run()
+
+        send.assert_called_once()
+
     @override_settings(
         FILENAME_FORMAT=None,
         TIME_ZONE="America/Chicago",
@@ -750,7 +909,7 @@ class TestConsumer(
         FILENAME_FORMAT="{correspondent}/{title}",
         ARCHIVE_FILE_GENERATION="always",
     )
-    @mock.patch("documents.signals.handlers.generate_unique_filename")
+    @mock.patch("documents.file_moves.generate_unique_filename")
     def testFilenameHandlingUnstableFormat(self, m) -> None:
         filenames = ["this", "that", "now this", "i cannot decide"]
 

src/documents/tests/test_file_handling.py

@@ -92,7 +92,7 @@ class TestFileHandling(DirectoriesMixin, FileSystemAssertsMixin, TestCase):
         # Ensure that filename is properly generated
         self.assertEqual(document.filename, Path("none/none.pdf"))
 
-        document.save()
+        document.save(write_file_paths=True)
 
         # test that creating dirs for the source_path creates the correct directory
         create_source_path_directory(document.source_path)
@@ -190,7 +190,7 @@ class TestFileHandling(DirectoriesMixin, FileSystemAssertsMixin, TestCase):
 
         # Set a correspondent and save the document
         document.correspondent = Correspondent.objects.get_or_create(name="test")[0]
-        document.save()
+        document.save(write_file_paths=True)
 
         # Check proper handling of files
         self.assertIsFile(
@@ -232,7 +232,7 @@ class TestFileHandling(DirectoriesMixin, FileSystemAssertsMixin, TestCase):
             disable_auditlog(),
         ):
             m.side_effect = DatabaseError()
-            document.save()
+            document.save(write_file_paths=True)
 
             # Check proper handling of files
             self.assertIsFile(document.source_path)
@@ -336,7 +336,7 @@ class TestFileHandling(DirectoriesMixin, FileSystemAssertsMixin, TestCase):
 
         # Ensure that filename is properly generated
         document.filename = generate_filename(document)
-        document.save()
+        document.save(write_file_paths=True)
         self.assertEqual(document.filename, "none/none.pdf")
 
         create_source_path_directory(document.source_path)
@@ -362,7 +362,7 @@ class TestFileHandling(DirectoriesMixin, FileSystemAssertsMixin, TestCase):
 
         # Ensure that filename is properly generated
         document.filename = generate_filename(document)
-        document.save()
+        document.save(write_file_paths=True)
         self.assertEqual(document.filename, "none/none.pdf")
 
         create_source_path_directory(document.source_path)
@@ -385,7 +385,7 @@ class TestFileHandling(DirectoriesMixin, FileSystemAssertsMixin, TestCase):
 
         document.save()
         document.filename = generate_filename(document)
-        document.save()
+        document.save(write_file_paths=True)
         create_source_path_directory(document.source_path)
         Path(document.source_path).touch()
         document.delete()
@@ -421,7 +421,7 @@ class TestFileHandling(DirectoriesMixin, FileSystemAssertsMixin, TestCase):
 
         # Set a correspondent and save the document
         document.correspondent = Correspondent.objects.get_or_create(name="test")[0]
-        document.save()
+        document.save(write_file_paths=True)
 
         # Check proper handling of files
         self.assertIsDir(settings.ORIGINALS_DIR / "test")
@@ -476,7 +476,7 @@ class TestFileHandling(DirectoriesMixin, FileSystemAssertsMixin, TestCase):
     def test_filename_relative(self) -> None:
         doc = Document.objects.create(title="doc1", mime_type="application/pdf")
         doc.filename = generate_filename(doc)
-        doc.save()
+        doc.save(write_file_paths=True)
 
         self.assertEqual(
             doc.source_path,
@@ -532,7 +532,7 @@ class TestFileHandling(DirectoriesMixin, FileSystemAssertsMixin, TestCase):
 
         # Ensure that filename is properly generated
         document.filename = generate_filename(document)
-        document.save()
+        document.save(write_file_paths=True)
         self.assertEqual(document.filename, "none/none/none.pdf")
         create_source_path_directory(document.source_path)
         Path(document.source_path).touch()
@@ -620,13 +620,13 @@ class TestFileHandling(DirectoriesMixin, FileSystemAssertsMixin, TestCase):
         Path(document.source_path).touch()
         Path(document2.source_path).touch()
         document.filename = "0000001.pdf"
-        document.save()
+        document.save(write_file_paths=True)
 
         self.assertIsFile(document.source_path)
         self.assertEqual(document.filename, "qwe.pdf")
 
         document2.filename = "0000002.pdf"
-        document2.save()
+        document2.save(write_file_paths=True)
 
         self.assertIsFile(document.source_path)
         self.assertEqual(document2.filename, "qwe_01.pdf")
@@ -656,19 +656,17 @@ class TestFileHandling(DirectoriesMixin, FileSystemAssertsMixin, TestCase):
         self.assertEqual(document2.filename, "qwe.pdf")
 
     @override_settings(FILENAME_FORMAT="{title}")
-    @mock.patch("documents.signals.handlers.Document.objects.filter")
-    @mock.patch("documents.signals.handlers.shutil.move")
+    @mock.patch("documents.file_moves.Document.global_objects.filter")
+    @mock.patch("documents.file_moves.shutil.move")
     def test_no_move_only_save(self, mock_move, mock_filter) -> None:
         """
         GIVEN:
             - A document with a filename
-            - The document is saved
-            - The filename is not changed
         WHEN:
-            - The document is saved
+            - The document is saved without a change that affects its path
         THEN:
-            - The document modified date is updated
-            - The document is not moved
+            - The document modified date is updated by the save itself
+            - The file is not moved and the mover writes nothing
         """
         with disable_auditlog():
             doc = Document.objects.create(
@@ -686,7 +684,7 @@ class TestFileHandling(DirectoriesMixin, FileSystemAssertsMixin, TestCase):
             doc.save()
             doc.refresh_from_db()
 
-            mock_filter.assert_called()
+            mock_filter.assert_not_called()
             self.assertNotEqual(original_modified, doc.modified)
             mock_move.assert_not_called()
 
@@ -694,7 +692,7 @@ class TestFileHandling(DirectoriesMixin, FileSystemAssertsMixin, TestCase):
         FILENAME_FORMAT="{{title}}_{{custom_fields|get_cf_value('test')}}",
         CELERY_TASK_ALWAYS_EAGER=True,
     )
-    @mock.patch("documents.signals.handlers.update_filename_and_move_files")
+    @mock.patch("documents.signals.handlers.move_files_for_document")
     def test_select_cf_updated(self, m) -> None:
         """
         GIVEN:
@@ -2045,7 +2043,7 @@ class TestCustomFieldFilenameUpdates(
         Path(self.doc.source_path).parent.mkdir(parents=True, exist_ok=True)
         Path(self.doc.source_path).touch()
 
-        with mock.patch("documents.signals.handlers.generate_unique_filename") as m:
+        with mock.patch("documents.file_moves.generate_unique_filename") as m:
             m.side_effect = generate_unique_filename
             self.cfi.value_text = "updated"
             self.cfi.save()
@@ -2067,7 +2065,7 @@ class TestCustomFieldFilenameUpdates(
         Path(self.doc.source_path).parent.mkdir(parents=True, exist_ok=True)
         Path(self.doc.source_path).touch()
 
-        with mock.patch("documents.signals.handlers.generate_unique_filename") as m:
+        with mock.patch("documents.file_moves.generate_unique_filename") as m:
             m.side_effect = generate_unique_filename
             self.cfi.value_text = "updated"
             self.cfi.save()
@@ -2145,7 +2143,7 @@ def test_refresh_failure_surfaces_database_error(
     GIVEN:
         - A document with a stored filename
     WHEN:
-        - Saving it triggers a filename update and refreshing it from the database fails
+        - Saving it triggers a filename update and loading it from the database fails
     THEN:
         - The database error is raised, not an unbound variable error from the rollback
     """
@@ -2153,8 +2151,8 @@ def test_refresh_failure_surfaces_database_error(
 
     with (
         mock.patch.object(
-            Document,
-            "refresh_from_db",
+            Document.global_objects,
+            "get",
             side_effect=DatabaseError("database is locked"),
         ),
         pytest.raises(DatabaseError, match="database is locked"),

src/documents/tests/test_file_moves.py

@@ -0,0 +1,944 @@
+from __future__ import annotations
+
+import datetime
+import hashlib
+import logging
+from pathlib import Path
+from typing import TYPE_CHECKING
+
+import pytest
+from asgiref.sync import async_to_sync
+from asgiref.sync import sync_to_async
+from celery.contrib.django.task import DjangoTask
+from django.db import transaction
+from django.http import HttpRequest
+from django.http import HttpResponse
+from django.test import RequestFactory
+from django.utils import timezone
+from django.utils.dateparse import parse_datetime
+from filelock import Timeout
+from rest_framework import status
+
+from documents.file_moves import _active_collector
+from documents.file_moves import defer_file_moves
+from documents.file_moves import flush_pending_file_moves
+from documents.file_moves import move_files_for_document
+from documents.file_moves import request_file_move
+from documents.models import CustomField
+from documents.models import Document
+from documents.models import Workflow
+from documents.models import WorkflowAction
+from documents.models import WorkflowTrigger
+from documents.tasks import bulk_update_documents
+from documents.tasks import check_scheduled_workflows
+from documents.tasks import empty_trash
+from paperless.celery import app as celery_app
+from paperless.middleware import FileMoveMiddleware
+from paperless_testing.factories import CorrespondentFactory
+from paperless_testing.factories import CustomFieldFactory
+from paperless_testing.factories import CustomFieldInstanceFactory
+from paperless_testing.factories import DocumentFactory
+from paperless_testing.factories import StoragePathFactory
+from paperless_testing.factories import TagFactory
+from paperless_testing.locks import record_media_lock_acquisitions
+
+if TYPE_CHECKING:
+    from collections.abc import Callable
+    from contextlib import AbstractContextManager
+
+    from pytest_django.fixtures import Settings
+    from pytest_mock import MockerFixture
+    from rest_framework.test import APIClient
+
+    from paperless_testing.dirs import PaperlessDirs
+
+    OnCommitCapture = Callable[..., AbstractContextManager[list[Callable[[], object]]]]
+
+CONTENT = b"original content"
+
+
+@pytest.fixture
+def title_format(settings: Settings) -> None:
+    settings.FILENAME_FORMAT = "{title}"
+
+
+@pytest.fixture
+def document(paperless_dirs: PaperlessDirs, title_format: None) -> Document:
+    """A document named after its title, with its original on disk."""
+    (paperless_dirs.originals_dir / "original.pdf").write_bytes(CONTENT)
+    return DocumentFactory(
+        title="original",
+        filename="original.pdf",
+        mime_type="application/pdf",
+        checksum=hashlib.sha256(CONTENT).hexdigest(),
+    )
+
+
+@celery_app.task(name="documents.tests.test_file_moves.rename_then_fail")
+def rename_then_fail(pk: int, originals_dir: str) -> None:
+    """Test-only task: rename the document, check nothing moved yet, then fail."""
+    doc = Document.objects.get(pk=pk)
+    doc.title = "renamed"
+    doc.save()
+    assert (Path(originals_dir) / "original.pdf").is_file()
+    raise RuntimeError("task failed after saving")
+
+
+@pytest.mark.django_db
+class TestMoveFilesForDocument:
+    def test_moves_using_fresh_db_state(
+        self,
+        document: Document,
+        paperless_dirs: PaperlessDirs,
+    ) -> None:
+        """
+        GIVEN:
+            - A document whose title changed in the DB without any signal
+        WHEN:
+            - move_files_for_document is called with its pk
+        THEN:
+            - The file and the DB filename follow the DB title
+            - The returned instance carries the final filename
+        """
+        Document.objects.filter(pk=document.pk).update(title="renamed")
+
+        moved = move_files_for_document(document.pk)
+
+        assert moved is not None
+        assert moved.filename == "renamed.pdf"
+        assert (paperless_dirs.originals_dir / "renamed.pdf").is_file()
+        assert not (paperless_dirs.originals_dir / "original.pdf").exists()
+        document.refresh_from_db()
+        assert document.filename == "renamed.pdf"
+
+    def test_missing_document_returns_none(self, paperless_dirs: PaperlessDirs) -> None:
+        """
+        GIVEN:
+            - A pk with no document row
+        WHEN:
+            - move_files_for_document is called
+        THEN:
+            - It returns None without raising
+        """
+        assert move_files_for_document(999_999) is None
+
+    def test_document_without_filename_skips_lock(
+        self,
+        paperless_dirs: PaperlessDirs,
+        title_format: None,
+    ) -> None:
+        """
+        GIVEN:
+            - A document still being consumed (no filename yet)
+        WHEN:
+            - move_files_for_document is called
+        THEN:
+            - It returns None, writes nothing, and never takes media.lock
+        """
+        doc = DocumentFactory(filename=None, mime_type="application/pdf")
+
+        with record_media_lock_acquisitions() as acquisitions:
+            assert move_files_for_document(doc.pk) is None
+
+        assert acquisitions == []
+        doc.refresh_from_db()
+        assert doc.filename is None
+
+    def test_no_change_does_not_write_modified(
+        self,
+        document: Document,
+    ) -> None:
+        """
+        GIVEN:
+            - A document already at its generated path, modified in the past
+        WHEN:
+            - move_files_for_document is called
+        THEN:
+            - modified is not written
+        """
+        past = timezone.now() - datetime.timedelta(days=1)
+        Document.objects.filter(pk=document.pk).update(modified=past)
+
+        move_files_for_document(document.pk)
+
+        document.refresh_from_db()
+        assert document.modified == past
+
+    def test_trashed_already_moved_document_gets_db_path_fixed(
+        self,
+        paperless_dirs: PaperlessDirs,
+        title_format: None,
+    ) -> None:
+        """
+        GIVEN:
+            - A trashed document whose DB filename is stale
+            - Its file already sits at the generated path (checksum matches)
+        WHEN:
+            - move_files_for_document is called
+        THEN:
+            - The DB filename is updated to where the file is
+        """
+        doc = DocumentFactory(
+            title="current",
+            filename="stale.pdf",
+            mime_type="application/pdf",
+            checksum=hashlib.sha256(CONTENT).hexdigest(),
+        )
+        assert Document.objects.get(pk=doc.pk).filename == "stale.pdf"
+        doc.delete()
+        (paperless_dirs.originals_dir / "current.pdf").write_bytes(CONTENT)
+
+        move_files_for_document(doc.pk)
+
+        assert Document.global_objects.get(pk=doc.pk).filename == "current.pdf"
+
+    def test_lock_timeout_leaves_consistent_state(
+        self,
+        document: Document,
+        paperless_dirs: PaperlessDirs,
+        mocker: MockerFixture,
+        caplog: pytest.LogCaptureFixture,
+    ) -> None:
+        """
+        GIVEN:
+            - A document whose title changed in the DB
+            - media.lock cannot be acquired
+        WHEN:
+            - move_files_for_document is called
+        THEN:
+            - An error is logged and None is returned
+            - The lock was requested with the 300 second timeout
+            - The DB filename and the file on disk are unchanged
+        """
+        Document.objects.filter(pk=document.pk).update(title="renamed")
+        lock = mocker.patch("documents.file_moves.media_lock")
+        lock.return_value.__enter__.side_effect = Timeout("media.lock")
+
+        with caplog.at_level(logging.ERROR, logger="paperless.handlers"):
+            assert move_files_for_document(document.pk) is None
+
+        assert "Timed out" in caplog.text
+        lock.assert_called_once_with(timeout=300)
+        document.refresh_from_db()
+        assert document.filename == "original.pdf"
+        assert (paperless_dirs.originals_dir / "original.pdf").is_file()
+
+
+@pytest.mark.django_db
+class TestRequestFileMoveWithoutScope:
+    def test_moves_immediately_and_syncs_only_file_columns(
+        self,
+        document: Document,
+        paperless_dirs: PaperlessDirs,
+    ) -> None:
+        """
+        GIVEN:
+            - No file move scope is open
+            - A caller instance with an unsaved in-memory change
+        WHEN:
+            - request_file_move is called after the title changed in the DB
+        THEN:
+            - The file moves immediately
+            - The caller's filename is updated
+            - The caller's unsaved change is kept
+        """
+        Document.objects.filter(pk=document.pk).update(title="renamed")
+        unsaved = CorrespondentFactory()
+        document.correspondent = unsaved
+
+        request_file_move(document)
+
+        assert (paperless_dirs.originals_dir / "renamed.pdf").is_file()
+        assert document.filename == "renamed.pdf"
+        assert document.correspondent == unsaved
+        assert document.title == "original"
+
+
+@pytest.mark.django_db
+class TestReceiver:
+    def test_reverse_m2m_does_not_request_a_move(
+        self,
+        document: Document,
+        mocker: MockerFixture,
+    ) -> None:
+        """
+        GIVEN:
+            - A tag
+        WHEN:
+            - A document is added through the tag's reverse relation
+        THEN:
+            - No move is requested with the tag as the instance
+        """
+        tag = TagFactory()
+        request = mocker.patch("documents.signals.handlers.request_file_move")
+
+        tag.documents.add(document)
+
+        request.assert_not_called()
+
+
+@pytest.mark.django_db
+class TestDeferFileMoves:
+    def test_moves_once_on_exit(
+        self,
+        document: Document,
+        paperless_dirs: PaperlessDirs,
+    ) -> None:
+        """
+        GIVEN:
+            - An open file move scope
+        WHEN:
+            - The document is saved with a new title inside the scope
+        THEN:
+            - Nothing moves while the scope is open
+            - On exit the file and DB filename follow the final title
+        """
+        with defer_file_moves():
+            document.title = "renamed"
+            document.save()
+            assert (paperless_dirs.originals_dir / "original.pdf").is_file()
+            assert Document.objects.get(pk=document.pk).filename == "original.pdf"
+
+        assert (paperless_dirs.originals_dir / "renamed.pdf").is_file()
+        assert not (paperless_dirs.originals_dir / "original.pdf").exists()
+        assert Document.objects.get(pk=document.pk).filename == "renamed.pdf"
+
+    def test_nested_scope_defers_to_outer(
+        self,
+        document: Document,
+        paperless_dirs: PaperlessDirs,
+    ) -> None:
+        """
+        GIVEN:
+            - A scope nested inside another
+        WHEN:
+            - The document is saved inside the inner scope and the inner scope exits
+        THEN:
+            - Nothing moves until the outer scope exits
+        """
+        with defer_file_moves():
+            with defer_file_moves():
+                document.title = "renamed"
+                document.save()
+            assert (paperless_dirs.originals_dir / "original.pdf").is_file()
+
+        assert (paperless_dirs.originals_dir / "renamed.pdf").is_file()
+
+    def test_closed_collector_moves_immediately(
+        self,
+        document: Document,
+        paperless_dirs: PaperlessDirs,
+    ) -> None:
+        """
+        GIVEN:
+            - A collector that has already been closed (work outliving its scope)
+        WHEN:
+            - A save asks for a move while that collector is still current
+        THEN:
+            - The move happens immediately instead of being lost
+        """
+        with defer_file_moves():
+            collector = _active_collector.get()
+        assert collector is not None
+        token = _active_collector.set(collector)
+        try:
+            document.title = "renamed"
+            document.save()
+        finally:
+            _active_collector.reset(token)
+
+        assert (paperless_dirs.originals_dir / "renamed.pdf").is_file()
+
+    def test_scope_closing_inside_atomic_flushes_on_commit(
+        self,
+        document: Document,
+        paperless_dirs: PaperlessDirs,
+        django_capture_on_commit_callbacks: OnCommitCapture,
+    ) -> None:
+        """
+        GIVEN:
+            - A scope opened and closed inside an atomic block
+        WHEN:
+            - The atomic block commits
+        THEN:
+            - The move runs only once the commit callbacks run
+        """
+        with django_capture_on_commit_callbacks(execute=True):
+            with transaction.atomic(), defer_file_moves():
+                document.title = "renamed"
+                document.save()
+            assert (paperless_dirs.originals_dir / "original.pdf").is_file()
+
+        assert (paperless_dirs.originals_dir / "renamed.pdf").is_file()
+
+    def test_scope_closing_inside_rolled_back_atomic_moves_nothing(
+        self,
+        document: Document,
+        paperless_dirs: PaperlessDirs,
+        django_capture_on_commit_callbacks: OnCommitCapture,
+    ) -> None:
+        """
+        GIVEN:
+            - A scope opened and closed inside an atomic block
+        WHEN:
+            - The atomic block rolls back
+        THEN:
+            - The title change is gone and the file is not moved
+        """
+
+        class Rollback(Exception):
+            pass
+
+        with django_capture_on_commit_callbacks(execute=True):
+            with pytest.raises(Rollback):
+                with transaction.atomic(), defer_file_moves():
+                    document.title = "renamed"
+                    document.save()
+                    raise Rollback
+
+        assert (paperless_dirs.originals_dir / "original.pdf").is_file()
+        assert not (paperless_dirs.originals_dir / "renamed.pdf").exists()
+        assert Document.objects.get(pk=document.pk).title == "original"
+
+    def test_flush_pending_moves_early(
+        self,
+        document: Document,
+        paperless_dirs: PaperlessDirs,
+    ) -> None:
+        """
+        GIVEN:
+            - An open scope with a pending move
+        WHEN:
+            - flush_pending_file_moves is called inside the scope
+        THEN:
+            - The move happens before the scope exits
+        """
+        with defer_file_moves():
+            document.title = "renamed"
+            document.save()
+            flush_pending_file_moves()
+            assert (paperless_dirs.originals_dir / "renamed.pdf").is_file()
+
+    def test_document_without_filename_in_scope_is_noop(
+        self,
+        paperless_dirs: PaperlessDirs,
+        title_format: None,
+    ) -> None:
+        """
+        GIVEN:
+            - A document still being consumed (no filename yet)
+        WHEN:
+            - It is saved inside a scope and the scope exits
+        THEN:
+            - Nothing is written, media.lock is never taken, nothing raises
+        """
+        doc = DocumentFactory(filename=None, mime_type="application/pdf")
+
+        with record_media_lock_acquisitions() as acquisitions:
+            with defer_file_moves():
+                doc.title = "renamed"
+                doc.save()
+
+        assert acquisitions == []
+        doc.refresh_from_db()
+        assert doc.filename is None
+
+    def test_mover_keeps_metadata_changed_after_save(
+        self,
+        document: Document,
+        paperless_dirs: PaperlessDirs,
+        settings: Settings,
+    ) -> None:
+        """
+        GIVEN:
+            - A filename format using correspondent and title
+            - A scope in which the title is saved
+        WHEN:
+            - Another writer changes the correspondent before the scope exits
+        THEN:
+            - The correspondent change survives the flush
+            - The final path reflects both changes
+        """
+        settings.FILENAME_FORMAT = "{correspondent}/{title}"
+        correspondent = CorrespondentFactory(name="acme")
+
+        with defer_file_moves():
+            document.title = "renamed"
+            document.save()
+            Document.objects.filter(pk=document.pk).update(correspondent=correspondent)
+
+        document.refresh_from_db()
+        assert document.correspondent == correspondent
+        assert document.filename == "acme/renamed.pdf"
+        assert (paperless_dirs.originals_dir / "acme" / "renamed.pdf").is_file()
+
+    def test_one_failure_does_not_stop_the_flush(
+        self,
+        paperless_dirs: PaperlessDirs,
+        title_format: None,
+        mocker: MockerFixture,
+    ) -> None:
+        """
+        GIVEN:
+            - Two documents saved in one scope
+            - Moving the first raises an unexpected error
+        WHEN:
+            - The scope exits
+        THEN:
+            - The scope does not raise
+            - The second document is still moved
+        """
+        (paperless_dirs.originals_dir / "a.pdf").write_bytes(CONTENT)
+        first = DocumentFactory(
+            filename="a.pdf",
+            title="a",
+            mime_type="application/pdf",
+        )
+        (paperless_dirs.originals_dir / "b.pdf").write_bytes(CONTENT)
+        second = DocumentFactory(
+            filename="b.pdf",
+            title="b",
+            mime_type="application/pdf",
+        )
+
+        real_move = move_files_for_document
+
+        def failing_for_first(pk: int) -> Document | None:
+            if pk == first.pk:
+                raise RuntimeError("boom")
+            return real_move(pk)
+
+        mocker.patch(
+            "documents.file_moves.move_files_for_document",
+            side_effect=failing_for_first,
+        )
+
+        with defer_file_moves():
+            first.title = "a2"
+            first.save()
+            second.title = "b2"
+            second.save()
+
+        assert (paperless_dirs.originals_dir / "b2.pdf").is_file()
+        assert (paperless_dirs.originals_dir / "a.pdf").is_file()
+
+
+@pytest.mark.django_db
+class TestFileMoveMiddleware:
+    def test_sync_mode_flushes_before_returning(
+        self,
+        document: Document,
+        paperless_dirs: PaperlessDirs,
+    ) -> None:
+        """
+        GIVEN:
+            - The middleware wrapping a sync view that renames the document
+        WHEN:
+            - A request passes through it
+        THEN:
+            - Nothing moves while the view runs
+            - The file is at its final path when the middleware returns
+        """
+
+        def view(request: HttpRequest) -> HttpResponse:
+            document.title = "renamed"
+            document.save()
+            assert (paperless_dirs.originals_dir / "original.pdf").is_file()
+            return HttpResponse()
+
+        middleware = FileMoveMiddleware(view)
+        assert middleware.async_mode is False
+
+        middleware(RequestFactory().get("/"))
+
+        assert (paperless_dirs.originals_dir / "renamed.pdf").is_file()
+
+    def test_async_mode_flushes_after_await(
+        self,
+        document: Document,
+        paperless_dirs: PaperlessDirs,
+    ) -> None:
+        """
+        GIVEN:
+            - The middleware wrapping an async view that saves through
+              sync_to_async
+        WHEN:
+            - A request passes through it
+        THEN:
+            - Nothing moves while the view runs
+            - The file is at its final path after the await, so the collector
+              survived the context copy into the sync thread
+        """
+
+        def rename() -> None:
+            document.title = "renamed"
+            document.save()
+
+        async def view(request: HttpRequest) -> HttpResponse:
+            await sync_to_async(rename)()
+            assert (paperless_dirs.originals_dir / "original.pdf").is_file()
+            return HttpResponse()
+
+        middleware = FileMoveMiddleware(view)
+        assert middleware.async_mode is True
+
+        async_to_sync(middleware)(RequestFactory().get("/"))
+
+        assert (paperless_dirs.originals_dir / "renamed.pdf").is_file()
+
+    def test_view_raising_after_save_still_moves(
+        self,
+        document: Document,
+        paperless_dirs: PaperlessDirs,
+    ) -> None:
+        """
+        GIVEN:
+            - A view that saves a rename and then raises
+        WHEN:
+            - A request passes through the middleware
+        THEN:
+            - The exception propagates
+            - The committed rename is still moved
+        """
+
+        def view(request: HttpRequest) -> HttpResponse:
+            document.title = "renamed"
+            document.save()
+            raise RuntimeError("view failed")
+
+        with pytest.raises(RuntimeError, match="view failed"):
+            FileMoveMiddleware(view)(RequestFactory().get("/"))
+
+        assert (paperless_dirs.originals_dir / "renamed.pdf").is_file()
+
+
+@pytest.mark.django_db
+class TestFileMovesThroughApi:
+    def test_media_lock_never_taken_inside_a_transaction(
+        self,
+        admin_client: APIClient,
+        document: Document,
+    ) -> None:
+        """
+        GIVEN:
+            - A document and a tag
+        WHEN:
+            - A tag + title PATCH and a note POST are made through the API
+        THEN:
+            - media.lock is acquired at least once
+            - media.lock is never first acquired while a transaction is open
+        """
+        tag = TagFactory()
+
+        with record_media_lock_acquisitions() as acquisitions:
+            response = admin_client.patch(
+                f"/api/documents/{document.pk}/",
+                {"tags": [tag.pk], "title": "renamed"},
+                format="json",
+            )
+            assert response.status_code == status.HTTP_200_OK
+            response = admin_client.post(
+                f"/api/documents/{document.pk}/notes/",
+                {"note": "a note"},
+                format="json",
+            )
+            assert response.status_code == status.HTTP_200_OK
+
+        assert acquisitions
+        assert not any(a.in_user_transaction for a in acquisitions)
+
+    def test_patch_moves_to_final_path(
+        self,
+        admin_client: APIClient,
+        document: Document,
+        paperless_dirs: PaperlessDirs,
+    ) -> None:
+        """
+        GIVEN:
+            - A document named after its title
+        WHEN:
+            - Its title and tags are changed through the API
+        THEN:
+            - The file is at the final path when the response returns
+        """
+        response = admin_client.patch(
+            f"/api/documents/{document.pk}/",
+            {"tags": [TagFactory().pk], "title": "renamed"},
+            format="json",
+        )
+
+        assert response.status_code == status.HTTP_200_OK
+        assert (paperless_dirs.originals_dir / "renamed.pdf").is_file()
+        assert Document.objects.get(pk=document.pk).filename == "renamed.pdf"
+
+    def test_patch_response_reports_post_move_modified(
+        self,
+        admin_client: APIClient,
+        document: Document,
+        paperless_dirs: PaperlessDirs,
+    ) -> None:
+        """
+        GIVEN:
+            - A document named after its title
+        WHEN:
+            - Its title is changed through the API, causing a rename
+        THEN:
+            - The modified time in the response equals the stored value,
+              which the file move bumped after the edit was saved
+        """
+        response = admin_client.patch(
+            f"/api/documents/{document.pk}/",
+            {"title": "renamed"},
+            format="json",
+        )
+
+        assert response.status_code == status.HTTP_200_OK
+        stored = Document.objects.get(pk=document.pk)
+        assert stored.filename == "renamed.pdf"
+        assert parse_datetime(response.json()["modified"]) == stored.modified
+
+    def test_trash_restore_keeps_db_and_disk_consistent(
+        self,
+        admin_client: APIClient,
+        document: Document,
+    ) -> None:
+        """
+        GIVEN:
+            - A document moved to the trash
+        WHEN:
+            - It is restored through the trash API
+        THEN:
+            - Its DB filename points at a file that exists
+        """
+        admin_client.delete(f"/api/documents/{document.pk}/")
+
+        response = admin_client.post(
+            "/api/trash/",
+            {"action": "restore", "documents": [document.pk]},
+            format="json",
+        )
+
+        assert response.status_code == status.HTTP_200_OK
+        restored = Document.objects.get(pk=document.pk)
+        assert restored.source_path.is_file()
+
+    def test_root_storage_path_change_moves_versions(
+        self,
+        admin_client: APIClient,
+        document: Document,
+        paperless_dirs: PaperlessDirs,
+    ) -> None:
+        """
+        GIVEN:
+            - A root document with one version, both with files on disk
+        WHEN:
+            - The root's storage path is changed through the API
+        THEN:
+            - The root and version files both follow the storage path
+        """
+        (paperless_dirs.originals_dir / "version.pdf").write_bytes(CONTENT)
+        version = DocumentFactory(
+            root_document=document,
+            title="version",
+            filename="version.pdf",
+            mime_type="application/pdf",
+            checksum=hashlib.sha256(CONTENT).hexdigest(),
+        )
+        storage_path = StoragePathFactory(path="stored/{{ title }}")
+
+        response = admin_client.patch(
+            f"/api/documents/{document.pk}/",
+            {"storage_path": storage_path.pk},
+            format="json",
+        )
+
+        assert response.status_code == status.HTTP_200_OK
+        document.refresh_from_db()
+        version.refresh_from_db()
+        assert document.filename.startswith("stored/")
+        assert version.filename.startswith("stored/")
+        assert document.source_path.is_file()
+        assert version.source_path.is_file()
+
+    def test_custom_field_change_moves_to_final_path(
+        self,
+        admin_client: APIClient,
+        document: Document,
+        paperless_dirs: PaperlessDirs,
+        settings: Settings,
+    ) -> None:
+        """
+        GIVEN:
+            - A filename format that uses a text custom field
+            - A document with that field set
+        WHEN:
+            - The field value is changed through the API
+        THEN:
+            - The file is at the path for the new value
+        """
+        settings.FILENAME_FORMAT = (
+            "{{ custom_fields|get_cf_value('flavor') }}/{{ title }}"
+        )
+        field = CustomFieldFactory(
+            name="flavor",
+            data_type=CustomField.FieldDataType.STRING,
+        )
+        CustomFieldInstanceFactory(field=field, document=document, value_text="apple")
+
+        response = admin_client.patch(
+            f"/api/documents/{document.pk}/",
+            {"custom_fields": [{"field": field.pk, "value": "cherry"}]},
+            format="json",
+        )
+
+        assert response.status_code == status.HTTP_200_OK
+        document.refresh_from_db()
+        assert document.filename == "cherry/original.pdf"
+        assert (paperless_dirs.originals_dir / "cherry" / "original.pdf").is_file()
+
+
+@pytest.mark.django_db
+class TestCeleryTaskScope:
+    def test_task_class_opens_scope(self) -> None:
+        """
+        GIVEN:
+            - The paperless Celery app
+        WHEN:
+            - A registered task is inspected
+        THEN:
+            - It runs through FileMoveScopedTask
+            - It keeps Celery's DjangoTask API, such as delay_on_commit
+        """
+        from paperless.celery import FileMoveScopedTask
+
+        assert isinstance(bulk_update_documents, FileMoveScopedTask)
+        assert isinstance(bulk_update_documents, DjangoTask)
+
+    def test_failing_task_still_moves_committed_change(
+        self,
+        document: Document,
+        paperless_dirs: PaperlessDirs,
+    ) -> None:
+        """
+        GIVEN:
+            - A task that renames a document, then raises
+        WHEN:
+            - The task is applied
+        THEN:
+            - The task fails with its own error
+            - Nothing moved while the task body ran
+            - The committed rename is moved when the task ends
+        """
+        result = rename_then_fail.apply(
+            args=[document.pk, str(paperless_dirs.originals_dir)],
+        )
+
+        assert result.failed()
+        assert isinstance(result.result, RuntimeError)
+        assert (paperless_dirs.originals_dir / "renamed.pdf").is_file()
+
+    def test_bulk_update_moves_each_document(
+        self,
+        document: Document,
+        paperless_dirs: PaperlessDirs,
+        mocker: MockerFixture,
+    ) -> None:
+        """
+        GIVEN:
+            - Two documents whose titles changed in the DB without signals
+        WHEN:
+            - bulk_update_documents runs for both
+        THEN:
+            - Each document's file follows its new title
+        """
+        mocker.patch("documents.search.get_backend")
+        (paperless_dirs.originals_dir / "second.pdf").write_bytes(CONTENT)
+        second = DocumentFactory(
+            title="second",
+            filename="second.pdf",
+            mime_type="application/pdf",
+            checksum=hashlib.sha256(CONTENT).hexdigest(),
+        )
+        Document.objects.filter(pk=document.pk).update(title="first-renamed")
+        Document.objects.filter(pk=second.pk).update(title="second-renamed")
+
+        bulk_update_documents.apply(args=[[document.pk, second.pk]])
+
+        assert (paperless_dirs.originals_dir / "first-renamed.pdf").is_file()
+        assert (paperless_dirs.originals_dir / "second-renamed.pdf").is_file()
+
+    def test_scheduled_workflows_move_each_document(
+        self,
+        document: Document,
+        paperless_dirs: PaperlessDirs,
+        mocker: MockerFixture,
+    ) -> None:
+        """
+        GIVEN:
+            - Two documents matched by a scheduled workflow, where running the
+              workflow renames the document
+        WHEN:
+            - check_scheduled_workflows runs
+        THEN:
+            - The first document's file is already moved when the workflow
+              runs for the second, not only when the task ends
+        """
+        (paperless_dirs.originals_dir / "second.pdf").write_bytes(CONTENT)
+        second = DocumentFactory(
+            title="second",
+            filename="second.pdf",
+            mime_type="application/pdf",
+            checksum=hashlib.sha256(CONTENT).hexdigest(),
+        )
+        trigger = WorkflowTrigger.objects.create(
+            type=WorkflowTrigger.WorkflowTriggerType.SCHEDULED,
+            schedule_offset_days=0,
+            schedule_date_field=WorkflowTrigger.ScheduleDateField.ADDED,
+        )
+        workflow = Workflow.objects.create(name="scheduled", order=0)
+        workflow.triggers.add(trigger)
+        workflow.actions.add(WorkflowAction.objects.create(assign_title="unused"))
+
+        seen: dict[int, bool] = {}
+
+        def rename(*, document: Document, **kwargs: object) -> None:
+            seen[document.pk] = (paperless_dirs.originals_dir / "first.pdf").is_file()
+            document.title = "first" if document.pk == first_pk else "other"
+            document.save()
+
+        first_pk = min(document.pk, second.pk)
+        mocker.patch("documents.tasks.run_workflows", side_effect=rename)
+        mocker.patch("documents.tasks.send_websocket_document_updated")
+
+        check_scheduled_workflows.apply()
+
+        second_pk = max(document.pk, second.pk)
+        assert seen[first_pk] is False
+        assert seen[second_pk] is True
+
+
+@pytest.mark.django_db
+class TestEmptyTrashLockOrdering:
+    def test_hard_delete_takes_media_lock_outside_transaction(
+        self,
+        document: Document,
+        paperless_dirs: PaperlessDirs,
+    ) -> None:
+        """
+        GIVEN:
+            - A trashed document with its file on disk
+        WHEN:
+            - The trash is emptied for it
+        THEN:
+            - media.lock is never first acquired inside a transaction
+            - The document row and its file are gone
+        """
+        document.delete()
+
+        with record_media_lock_acquisitions() as acquisitions:
+            empty_trash([document.pk])
+
+        assert acquisitions
+        assert not any(a.in_user_transaction for a in acquisitions)
+        assert not Document.global_objects.filter(pk=document.pk).exists()
+        assert not (paperless_dirs.originals_dir / "original.pdf").exists()

src/documents/tests/test_management.py

@@ -18,11 +18,13 @@ if TYPE_CHECKING:
     from pytest_mock import MockerFixture
 
 from documents.file_handling import generate_filename
+from documents.file_moves import move_files_for_document
 from documents.models import Document
 from documents.tasks import update_document_content_maybe_archive_file
 from paperless_testing.assertions import FileSystemAssertsMixin
 from paperless_testing.dirs import DirectoriesMixin
 from paperless_testing.factories import DocumentFactory
+from paperless_testing.locks import record_media_lock_acquisitions
 from paperless_testing.samples import SIMPLE_DIGITAL_PDF
 
 sample_file: Path = SIMPLE_DIGITAL_PDF
@@ -63,6 +65,28 @@ class TestArchiver(DirectoriesMixin, FileSystemAssertsMixin, TestCase):
         self.assertTrue(filecmp.cmp(sample_file, doc.source_path))
         self.assertEqual(doc.archive_filename, "none/A.pdf")
 
+    def test_handle_document_takes_media_lock_outside_transaction(self) -> None:
+        """
+        GIVEN:
+            - A document whose archive version is being generated
+        WHEN:
+            - update_document_content_maybe_archive_file runs
+        THEN:
+            - media.lock is acquired
+            - media.lock is never first acquired while a transaction holds the
+              document row
+        """
+        doc = self.make_models()
+        shutil.copy(sample_file, Path(self.dirs.originals_dir) / f"{doc.id:07}.pdf")
+
+        with record_media_lock_acquisitions() as acquisitions:
+            update_document_content_maybe_archive_file(doc.pk)
+
+        self.assertTrue(acquisitions)
+        self.assertFalse(any(a.in_user_transaction for a in acquisitions))
+        doc.refresh_from_db()
+        self.assertIsFile(doc.archive_path)
+
     def test_unknown_mime_type(self) -> None:
         doc = self.make_models()
         doc.mime_type = "sdgfh"
@@ -236,7 +260,7 @@ class TestRenamer(DirectoriesMixin, FileSystemAssertsMixin, TestCase):
         doc = Document.objects.create(title="test", mime_type="image/jpeg")
         doc.filename = generate_filename(doc)
         doc.archive_filename = generate_filename(doc, archive_filename=True)
-        doc.save()
+        doc.save(write_file_paths=True)
 
         Path(doc.source_path).touch()
         Path(doc.archive_path).touch()
@@ -253,6 +277,42 @@ class TestRenamer(DirectoriesMixin, FileSystemAssertsMixin, TestCase):
         self.assertIsFile(doc2.source_path)
         self.assertIsFile(doc2.archive_path)
 
+    @override_settings(FILENAME_FORMAT="")
+    def test_rename_moves_each_document_by_direct_call(self) -> None:
+        """
+        GIVEN:
+            - Two documents whose filename format changed
+        WHEN:
+            - document_renamer runs
+        THEN:
+            - move_files_for_document is called once per document
+            - Both files are at their new paths
+        """
+        docs = []
+        for title in ("first", "second"):
+            doc = Document.objects.create(title=title, mime_type="image/jpeg")
+            doc.filename = generate_filename(doc)
+            doc.save(write_file_paths=True)
+            Path(doc.source_path).touch()
+            docs.append(doc)
+
+        with (
+            override_settings(FILENAME_FORMAT="{title}"),
+            mock.patch(
+                "documents.management.commands.document_renamer.move_files_for_document",
+                wraps=move_files_for_document,
+            ) as mover,
+        ):
+            call_command("document_renamer", skip_checks=True)
+
+        self.assertCountEqual(
+            [c.args[0] for c in mover.call_args_list],
+            [d.pk for d in docs],
+        )
+        for doc in docs:
+            doc.refresh_from_db()
+            self.assertIsFile(doc.source_path)
+
 
 @pytest.mark.management
 class TestCreateClassifier:

src/documents/tests/test_management_retagger.py

@@ -4,10 +4,13 @@ Tests for the document_retagger management command.
 
 from __future__ import annotations
 
+from typing import TYPE_CHECKING
+
 import pytest
 from django.core.management import call_command
 from django.core.management.base import CommandError
 
+from documents.file_moves import _active_collector
 from documents.models import Correspondent
 from documents.models import Document
 from documents.models import DocumentType
@@ -21,6 +24,12 @@ from paperless_testing.factories import DocumentTypeFactory
 from paperless_testing.factories import StoragePathFactory
 from paperless_testing.factories import TagFactory
 
+if TYPE_CHECKING:
+    from pytest_django.fixtures import Settings
+    from pytest_mock import MockerFixture
+
+    from paperless_testing.dirs import PaperlessDirs
+
 # ---------------------------------------------------------------------------
 # Module-level type aliases
 # ---------------------------------------------------------------------------
@@ -484,3 +493,69 @@ class TestRetaggerEdgeCases(DirectoriesMixin):
 
         assert d_first.tags.count() == 0
         assert d_unrelated.tags.count() == 2
+
+
+@pytest.mark.management
+@pytest.mark.django_db
+class TestRetaggerFileMoves:
+    def test_each_document_is_processed_in_its_own_scope(
+        self,
+        paperless_dirs: PaperlessDirs,
+        settings: Settings,
+        mocker: MockerFixture,
+    ) -> None:
+        """
+        GIVEN:
+            - Two documents and a filename format using the correspondent
+            - A classifier suggestion assigning a correspondent to each
+        WHEN:
+            - document_retagger runs with --correspondent
+        THEN:
+            - Each document is processed inside an open file move scope
+            - When the second document is processed, the first one's file has
+              already moved
+            - Both files end at their final paths
+        """
+        settings.FILENAME_FORMAT = "{correspondent}/{title}"
+        (paperless_dirs.originals_dir / "none").mkdir(exist_ok=True)
+        for title in ("first", "second"):
+            (paperless_dirs.originals_dir / "none" / f"{title}.pdf").write_bytes(b"x")
+            DocumentFactory(
+                title=title,
+                filename=f"none/{title}.pdf",
+                mime_type="application/pdf",
+            )
+        correspondent = CorrespondentFactory(name="acme")
+        processed: list[Document] = []
+        scope_open: list[bool] = []
+        seen_first_moved: list[bool] = []
+
+        def fake_set_correspondent(sender, document, **kwargs):
+            scope_open.append(_active_collector.get() is not None)
+            # Order-agnostic: on the second call, check the first one's file.
+            if processed:
+                first = processed[0]
+                seen_first_moved.append(
+                    (
+                        paperless_dirs.originals_dir / "acme" / f"{first.title}.pdf"
+                    ).is_file(),
+                )
+            processed.append(document)
+            document.correspondent = correspondent
+            document.save(update_fields=("correspondent",))
+            return correspondent
+
+        mocker.patch(
+            "documents.management.commands.document_retagger.load_classifier",
+        )
+        mocker.patch(
+            "documents.management.commands.document_retagger.set_correspondent",
+            side_effect=fake_set_correspondent,
+        )
+
+        call_command("document_retagger", "--correspondent", skip_checks=True)
+
+        assert scope_open == [True, True]
+        assert seen_first_moved == [True]
+        for title in ("first", "second"):
+            assert (paperless_dirs.originals_dir / "acme" / f"{title}.pdf").is_file()

src/documents/tests/test_media_lock.py

@@ -0,0 +1,121 @@
+from __future__ import annotations
+
+import threading
+from typing import TYPE_CHECKING
+
+import pytest
+from django.db import transaction
+from filelock import Timeout
+
+from documents.file_moves import in_user_transaction
+from documents.media_lock import media_lock
+from paperless_testing.locks import record_media_lock_acquisitions
+
+if TYPE_CHECKING:
+    from paperless_testing.dirs import PaperlessDirs
+
+
+@pytest.mark.usefixtures("paperless_dirs")
+class TestMediaLock:
+    def test_same_thread_reenters_without_blocking(self) -> None:
+        """
+        GIVEN:
+            - media.lock held by the current thread
+        WHEN:
+            - The same thread acquires it again with a zero timeout
+        THEN:
+            - The nested acquisition succeeds immediately
+        """
+        with media_lock(), media_lock(timeout=0):
+            pass
+
+    def test_other_thread_is_excluded(self) -> None:
+        """
+        GIVEN:
+            - media.lock held by the current thread
+        WHEN:
+            - Another thread tries to acquire it with a short timeout
+        THEN:
+            - The other thread times out
+            - After release, the other thread can acquire it
+        """
+        outcomes: list[str] = []
+
+        def try_acquire() -> None:
+            try:
+                with media_lock(timeout=0.2):
+                    outcomes.append("acquired")
+            except Timeout:
+                outcomes.append("timeout")
+
+        with media_lock():
+            worker = threading.Thread(target=try_acquire)
+            worker.start()
+            worker.join()
+
+        worker = threading.Thread(target=try_acquire)
+        worker.start()
+        worker.join()
+
+        assert outcomes == ["timeout", "acquired"]
+
+    def test_lock_file_is_the_configured_path(
+        self,
+        paperless_dirs: PaperlessDirs,
+    ) -> None:
+        """
+        GIVEN:
+            - The test's media directory layout
+        WHEN:
+            - media.lock is acquired
+        THEN:
+            - The lock file is created at settings.MEDIA_LOCK
+        """
+        with media_lock():
+            assert paperless_dirs.media_lock.exists()
+
+
+@pytest.mark.django_db
+class TestInUserTransaction:
+    def test_false_inside_test_case_wrapper(self) -> None:
+        """
+        GIVEN:
+            - The transaction pytest-django wraps around the test
+        WHEN:
+            - in_user_transaction is called
+        THEN:
+            - It returns False
+        """
+        assert in_user_transaction() is False
+
+    def test_true_inside_atomic(self) -> None:
+        """
+        GIVEN:
+            - An atomic block opened by code under test
+        WHEN:
+            - in_user_transaction is called inside it
+        THEN:
+            - It returns True
+        """
+        with transaction.atomic():
+            assert in_user_transaction() is True
+
+
+@pytest.mark.django_db
+@pytest.mark.usefixtures("paperless_dirs")
+class TestRecordMediaLockAcquisitions:
+    def test_records_only_first_acquisitions(self) -> None:
+        """
+        GIVEN:
+            - Nested media.lock acquisitions, the outer one inside a transaction
+        WHEN:
+            - They are recorded
+        THEN:
+            - Only the outer, first acquisition is recorded, flagged as inside
+              a transaction
+        """
+        with record_media_lock_acquisitions() as acquisitions:
+            with transaction.atomic(), media_lock(), media_lock():
+                pass
+
+        assert [a.in_user_transaction for a in acquisitions] == [True]

src/documents/tests/test_models.py

@@ -1,4 +1,7 @@
 import pytest
+from django.db import DatabaseError
+from django.db import models
+from django.db import transaction
 
 from documents.models import Correspondent
 from documents.models import Document
@@ -22,3 +25,101 @@ class TestDocument:
         assert Correspondent.objects.count() == 0
         assert Document.objects.count() == 1
         assert Document.objects.first().correspondent is None
+
+
+@pytest.mark.django_db
+@pytest.mark.usefixtures("paperless_dirs")
+class TestDocumentFilePathWrites:
+    def test_full_save_does_not_write_file_paths(self) -> None:
+        """
+        GIVEN:
+            - A document whose filename columns were changed in the DB after
+              an instance was loaded
+        WHEN:
+            - The stale instance is saved without update_fields
+        THEN:
+            - The DB filename columns keep the newer values
+            - Other fields from the stale instance are written
+        """
+        doc = DocumentFactory(filename="old.pdf", archive_filename="old.pdf")
+        stale = Document.objects.get(pk=doc.pk)
+        Document.objects.filter(pk=doc.pk).update(
+            filename="new.pdf",
+            archive_filename="new-archive.pdf",
+        )
+
+        stale.title = "changed"
+        stale.save()
+
+        doc.refresh_from_db()
+        assert doc.filename == "new.pdf"
+        assert doc.archive_filename == "new-archive.pdf"
+        assert doc.title == "changed"
+
+    @pytest.mark.parametrize(
+        "save_kwargs",
+        [
+            pytest.param({"write_file_paths": True}, id="write-file-paths"),
+            pytest.param(
+                {"update_fields": ["filename", "archive_filename"]},
+                id="explicit-update-fields",
+            ),
+        ],
+    )
+    def test_opt_in_writes_file_paths(self, save_kwargs: dict[str, object]) -> None:
+        """
+        GIVEN:
+            - An existing document
+        WHEN:
+            - Its filename columns are changed in memory and saved with an
+              explicit opt in
+        THEN:
+            - The DB filename columns are updated
+        """
+        doc = DocumentFactory(filename="old.pdf", archive_filename="old.pdf")
+
+        doc.filename = "new.pdf"
+        doc.archive_filename = "new.pdf"
+        doc.save(**save_kwargs)
+
+        doc.refresh_from_db()
+        assert doc.filename == "new.pdf"
+        assert doc.archive_filename == "new.pdf"
+
+    def test_creation_writes_file_paths(self) -> None:
+        """
+        GIVEN:
+            - A new, unsaved document with filename columns set
+        WHEN:
+            - It is saved
+        THEN:
+            - The filename columns are written
+        """
+        doc = DocumentFactory.build(filename="a.pdf", archive_filename="b.pdf")
+
+        doc.save()
+
+        doc.refresh_from_db()
+        assert doc.filename == "a.pdf"
+        assert doc.archive_filename == "b.pdf"
+
+    def test_full_save_of_deleted_row_raises(self) -> None:
+        """
+        GIVEN:
+            - An instance whose row was hard-deleted elsewhere
+        WHEN:
+            - The instance is saved without update_fields
+        THEN:
+            - A DatabaseError is raised instead of re-inserting the row
+        """
+        doc = DocumentFactory()
+        stale = Document.objects.get(pk=doc.pk)
+        # A plain QuerySet bypasses the soft-delete manager, so the row is
+        # really gone (as after emptying the trash).
+        models.QuerySet(model=Document).filter(pk=doc.pk).delete()
+
+        # The savepoint keeps the failed UPDATE from poisoning the test transaction.
+        with pytest.raises(DatabaseError), transaction.atomic():
+            stale.save()
+
+        assert not Document.global_objects.filter(pk=doc.pk).exists()

src/documents/tests/test_sanity_check.py

@@ -132,7 +132,7 @@ class TestCheckSanityOriginal:
 class TestCheckSanityArchive:
     def test_checksum_without_filename(self, sample_doc: Document) -> None:
         sample_doc.archive_filename = None
-        sample_doc.save()
+        sample_doc.save(write_file_paths=True)
         messages = check_sanity()
         assert messages.has_error
         assert any(
@@ -185,7 +185,7 @@ class TestCheckSanityArchive:
         Path(sample_doc.archive_path).unlink()
         sample_doc.archive_checksum = None
         sample_doc.archive_filename = None
-        sample_doc.save()
+        sample_doc.save(write_file_paths=True)
         messages = check_sanity()
         assert not messages.has_error
 

src/documents/tests/test_share_link_bundles.py

@@ -366,7 +366,7 @@ class ShareLinkBundleBuildTaskTests(DirectoriesMixin, APITestCase):
     def _write_document_file(self, *, archive: bool, content: bytes) -> Path:
         if archive:
             self.document.archive_filename = f"{self.document.pk:07}.pdf"
-            self.document.save()
+            self.document.save(write_file_paths=True)
             path = self.document.archive_path
             assert path is not None
         else:

src/documents/views.py

@@ -126,6 +126,7 @@ from documents.data_models import ConsumableDocument
 from documents.data_models import DocumentMetadataOverrides
 from documents.data_models import DocumentSource
 from documents.file_handling import format_filename
+from documents.file_moves import flush_pending_file_moves
 from documents.filters import CorrespondentFilterSet
 from documents.filters import CustomFieldFilterSet
 from documents.filters import DocumentFilterSet
@@ -1264,6 +1265,8 @@ class DocumentViewSet(
         )
         serializer.is_valid(raise_exception=True)
         self.perform_update(serializer)
+        # The response must reflect the post-move filename and modified time
+        flush_pending_file_moves()
 
         if content_updated and content_doc.id != root_doc.id:
             content_doc.content = (

src/paperless/celery.py

@@ -1,6 +1,7 @@
 import os
 
 from celery import Celery
+from celery.contrib.django.task import DjangoTask
 from celery.signals import worker_process_init
 from kombu.serialization import register
 
@@ -26,7 +27,28 @@ register(
     content_encoding="binary",
 )
 
-app = Celery("paperless")
+
+class FileMoveScopedTask(DjangoTask):
+    """
+    Run every task inside a document file move scope, so each document a task
+    changes is moved once when the task ends, after its writes have committed.
+    The scope also flushes when the task raises, for writes that committed
+    before the failure.
+
+    Subclasses DjangoTask because a custom task_cls replaces the Django fixup's
+    default, and delay_on_commit lives there.
+    """
+
+    def __call__(self, *args, **kwargs):
+        # Function-local: this module is imported by paperless/__init__.py
+        # before Django apps are loaded.
+        from documents.file_moves import defer_file_moves
+
+        with defer_file_moves():
+            return super().__call__(*args, **kwargs)
+
+
+app = Celery("paperless", task_cls=FileMoveScopedTask)
 
 # Using a string here means the worker doesn't have to serialize
 # the configuration object to child processes.

src/paperless/middleware.py

@@ -1,3 +1,5 @@
+from asgiref.sync import iscoroutinefunction
+from asgiref.sync import markcoroutinefunction
 from compression_middleware.middleware import CompressionMiddleware
 from django.conf import settings
 
@@ -30,3 +32,35 @@ class ApiVersionMiddleware:
             response["X-Version"] = version.__full_version_str__
 
         return response
+
+
+class FileMoveMiddleware:
+    """
+    Open a document file move scope around each request, so every document the
+    request changes is moved once, after its writes have committed, before the
+    response is returned.
+    """
+
+    sync_capable = True
+    async_capable = True
+
+    def __init__(self, get_response) -> None:
+        self.get_response = get_response
+        self.async_mode = iscoroutinefunction(get_response)
+        if self.async_mode:
+            markcoroutinefunction(self)
+
+    def __call__(self, request):
+        if self.async_mode:
+            return self.__acall__(request)
+
+        from documents.file_moves import defer_file_moves
+
+        with defer_file_moves():
+            return self.get_response(request)
+
+    async def __acall__(self, request):
+        from documents.file_moves import adefer_file_moves
+
+        async with adefer_file_moves():
+            return await self.get_response(request)

src/paperless/settings/__init__.py

@@ -189,6 +189,7 @@ if DEBUG:
 
 MIDDLEWARE = [
     "django.middleware.security.SecurityMiddleware",
+    "paperless.middleware.FileMoveMiddleware",
     "whitenoise.middleware.WhiteNoiseMiddleware",
     "django.contrib.sessions.middleware.SessionMiddleware",
     "corsheaders.middleware.CorsMiddleware",

src/paperless_testing/factories.py

@@ -8,6 +8,8 @@ import factory
 from django.contrib.auth.models import User
 
 from documents.models import Correspondent
+from documents.models import CustomField
+from documents.models import CustomFieldInstance
 from documents.models import Document
 from documents.models import DocumentBarcode
 from documents.models import DocumentType
@@ -70,6 +72,22 @@ class DocumentFactory(TypedModelFactory[Document]):
     storage_path = None
 
 
+class CustomFieldFactory(TypedModelFactory[CustomField]):
+    class Meta:
+        model = CustomField
+
+    name = factory.Sequence(lambda n: f"field {n}")
+    data_type = CustomField.FieldDataType.STRING
+
+
+class CustomFieldInstanceFactory(TypedModelFactory[CustomFieldInstance]):
+    class Meta:
+        model = CustomFieldInstance
+
+    field = factory.SubFactory(CustomFieldFactory)
+    document = factory.SubFactory(DocumentFactory)
+
+
 class DocumentBarcodeFactory(TypedModelFactory[DocumentBarcode]):
     class Meta:
         model = DocumentBarcode

src/paperless_testing/locks.py

@@ -0,0 +1,47 @@
+"""Record first acquisitions of the global media lock during a test."""
+
+from __future__ import annotations
+
+from contextlib import contextmanager
+from dataclasses import dataclass
+from pathlib import Path
+from typing import TYPE_CHECKING
+from unittest import mock
+
+if TYPE_CHECKING:
+    from collections.abc import Iterator
+
+
+@dataclass(frozen=True, slots=True)
+class MediaLockAcquisition:
+    """One first (non-reentrant) acquisition of settings.MEDIA_LOCK."""
+
+    in_user_transaction: bool
+
+
+@contextmanager
+def record_media_lock_acquisitions() -> Iterator[list[MediaLockAcquisition]]:
+    """
+    Record every first acquisition of settings.MEDIA_LOCK made in the block.
+
+    Reentrant acquisitions by a thread that already holds the lock never wait,
+    so they are not recorded. Each record notes whether a real (non test case)
+    transaction was open at that moment, which is the lock ordering hazard.
+    """
+    from django.conf import settings
+    from filelock import BaseFileLock
+
+    from documents.file_moves import in_user_transaction
+
+    records: list[MediaLockAcquisition] = []
+    original_acquire = BaseFileLock.acquire
+
+    def recording_acquire(self: BaseFileLock, *args: object, **kwargs: object):
+        if Path(self.lock_file) == Path(settings.MEDIA_LOCK) and not self.is_locked:
+            records.append(
+                MediaLockAcquisition(in_user_transaction=in_user_transaction()),
+            )
+        return original_acquire(self, *args, **kwargs)
+
+    with mock.patch.object(BaseFileLock, "acquire", recording_acquire):
+        yield records