Coverage for /pythoncovmergedfiles/medio/medio/usr/local/lib/python3.8/site-packages/joblib/_memmapping_reducer.py: 23%
250 statements
« prev ^ index » next coverage.py v7.3.2, created at 2023-12-12 06:31 +0000
1"""
2Reducer using memory mapping for numpy arrays
3"""
4# Author: Thomas Moreau <thomas.moreau.2010@gmail.com>
5# Copyright: 2017, Thomas Moreau
6# License: BSD 3 clause
8from mmap import mmap
9import errno
10import os
11import stat
12import threading
13import atexit
14import tempfile
15import time
16import warnings
17import weakref
18from uuid import uuid4
19from multiprocessing import util
21from pickle import whichmodule, loads, dumps, HIGHEST_PROTOCOL, PicklingError
23try:
24 WindowsError
25except NameError:
26 WindowsError = type(None)
28try:
29 import numpy as np
30 from numpy.lib.stride_tricks import as_strided
31except ImportError:
32 np = None
34from .numpy_pickle import dump, load, load_temporary_memmap
35from .backports import make_memmap
36from .disk import delete_folder
37from .externals.loky.backend import resource_tracker
39# Some system have a ramdisk mounted by default, we can use it instead of /tmp
40# as the default folder to dump big arrays to share with subprocesses.
41SYSTEM_SHARED_MEM_FS = '/dev/shm'
43# Minimal number of bytes available on SYSTEM_SHARED_MEM_FS to consider using
44# it as the default folder to dump big arrays to share with subprocesses.
45SYSTEM_SHARED_MEM_FS_MIN_SIZE = int(2e9)
47# Folder and file permissions to chmod temporary files generated by the
48# memmapping pool. Only the owner of the Python process can access the
49# temporary files and folder.
50FOLDER_PERMISSIONS = stat.S_IRUSR | stat.S_IWUSR | stat.S_IXUSR
51FILE_PERMISSIONS = stat.S_IRUSR | stat.S_IWUSR
53# Set used in joblib workers, referencing the filenames of temporary memmaps
54# created by joblib to speed up data communication. In child processes, we add
55# a finalizer to these memmaps that sends a maybe_unlink call to the
56# resource_tracker, in order to free main memory as fast as possible.
57JOBLIB_MMAPS = set()
60def _log_and_unlink(filename):
61 from .externals.loky.backend.resource_tracker import _resource_tracker
62 util.debug(
63 "[FINALIZER CALL] object mapping to {} about to be deleted,"
64 " decrementing the refcount of the file (pid: {})".format(
65 os.path.basename(filename), os.getpid()))
66 _resource_tracker.maybe_unlink(filename, "file")
69def add_maybe_unlink_finalizer(memmap):
70 util.debug(
71 "[FINALIZER ADD] adding finalizer to {} (id {}, filename {}, pid {})"
72 "".format(type(memmap), id(memmap), os.path.basename(memmap.filename),
73 os.getpid()))
74 weakref.finalize(memmap, _log_and_unlink, memmap.filename)
77def unlink_file(filename):
78 """Wrapper around os.unlink with a retry mechanism.
80 The retry mechanism has been implemented primarily to overcome a race
81 condition happening during the finalizer of a np.memmap: when a process
82 holding the last reference to a mmap-backed np.memmap/np.array is about to
83 delete this array (and close the reference), it sends a maybe_unlink
84 request to the resource_tracker. This request can be processed faster than
85 it takes for the last reference of the memmap to be closed, yielding (on
86 Windows) a PermissionError in the resource_tracker loop.
87 """
88 NUM_RETRIES = 10
89 for retry_no in range(1, NUM_RETRIES + 1):
90 try:
91 os.unlink(filename)
92 break
93 except PermissionError:
94 util.debug(
95 '[ResourceTracker] tried to unlink {}, got '
96 'PermissionError'.format(filename)
97 )
98 if retry_no == NUM_RETRIES:
99 raise
100 else:
101 time.sleep(.2)
102 except FileNotFoundError:
103 # In case of a race condition when deleting the temporary folder,
104 # avoid noisy FileNotFoundError exception in the resource tracker.
105 pass
108resource_tracker._CLEANUP_FUNCS['file'] = unlink_file
111class _WeakArrayKeyMap:
112 """A variant of weakref.WeakKeyDictionary for unhashable numpy arrays.
114 This datastructure will be used with numpy arrays as obj keys, therefore we
115 do not use the __get__ / __set__ methods to avoid any conflict with the
116 numpy fancy indexing syntax.
117 """
119 def __init__(self):
120 self._data = {}
122 def get(self, obj):
123 ref, val = self._data[id(obj)]
124 if ref() is not obj:
125 # In case of race condition with on_destroy: could never be
126 # triggered by the joblib tests with CPython.
127 raise KeyError(obj)
128 return val
130 def set(self, obj, value):
131 key = id(obj)
132 try:
133 ref, _ = self._data[key]
134 if ref() is not obj:
135 # In case of race condition with on_destroy: could never be
136 # triggered by the joblib tests with CPython.
137 raise KeyError(obj)
138 except KeyError:
139 # Insert the new entry in the mapping along with a weakref
140 # callback to automatically delete the entry from the mapping
141 # as soon as the object used as key is garbage collected.
142 def on_destroy(_):
143 del self._data[key]
144 ref = weakref.ref(obj, on_destroy)
145 self._data[key] = ref, value
147 def __getstate__(self):
148 raise PicklingError("_WeakArrayKeyMap is not pickleable")
151###############################################################################
152# Support for efficient transient pickling of numpy data structures
155def _get_backing_memmap(a):
156 """Recursively look up the original np.memmap instance base if any."""
157 b = getattr(a, 'base', None)
158 if b is None:
159 # TODO: check scipy sparse datastructure if scipy is installed
160 # a nor its descendants do not have a memmap base
161 return None
163 elif isinstance(b, mmap):
164 # a is already a real memmap instance.
165 return a
167 else:
168 # Recursive exploration of the base ancestry
169 return _get_backing_memmap(b)
172def _get_temp_dir(pool_folder_name, temp_folder=None):
173 """Get the full path to a subfolder inside the temporary folder.
175 Parameters
176 ----------
177 pool_folder_name : str
178 Sub-folder name used for the serialization of a pool instance.
180 temp_folder: str, optional
181 Folder to be used by the pool for memmapping large arrays
182 for sharing memory with worker processes. If None, this will try in
183 order:
185 - a folder pointed by the JOBLIB_TEMP_FOLDER environment
186 variable,
187 - /dev/shm if the folder exists and is writable: this is a
188 RAMdisk filesystem available by default on modern Linux
189 distributions,
190 - the default system temporary folder that can be
191 overridden with TMP, TMPDIR or TEMP environment
192 variables, typically /tmp under Unix operating systems.
194 Returns
195 -------
196 pool_folder : str
197 full path to the temporary folder
198 use_shared_mem : bool
199 whether the temporary folder is written to the system shared memory
200 folder or some other temporary folder.
201 """
202 use_shared_mem = False
203 if temp_folder is None:
204 temp_folder = os.environ.get('JOBLIB_TEMP_FOLDER', None)
205 if temp_folder is None:
206 if os.path.exists(SYSTEM_SHARED_MEM_FS) and hasattr(os, 'statvfs'):
207 try:
208 shm_stats = os.statvfs(SYSTEM_SHARED_MEM_FS)
209 available_nbytes = shm_stats.f_bsize * shm_stats.f_bavail
210 if available_nbytes > SYSTEM_SHARED_MEM_FS_MIN_SIZE:
211 # Try to see if we have write access to the shared mem
212 # folder only if it is reasonably large (that is 2GB or
213 # more).
214 temp_folder = SYSTEM_SHARED_MEM_FS
215 pool_folder = os.path.join(temp_folder, pool_folder_name)
216 if not os.path.exists(pool_folder):
217 os.makedirs(pool_folder)
218 use_shared_mem = True
219 except (IOError, OSError):
220 # Missing rights in the /dev/shm partition, fallback to regular
221 # temp folder.
222 temp_folder = None
223 if temp_folder is None:
224 # Fallback to the default tmp folder, typically /tmp
225 temp_folder = tempfile.gettempdir()
226 temp_folder = os.path.abspath(os.path.expanduser(temp_folder))
227 pool_folder = os.path.join(temp_folder, pool_folder_name)
228 return pool_folder, use_shared_mem
231def has_shareable_memory(a):
232 """Return True if a is backed by some mmap buffer directly or not."""
233 return _get_backing_memmap(a) is not None
236def _strided_from_memmap(filename, dtype, mode, offset, order, shape, strides,
237 total_buffer_len, unlink_on_gc_collect):
238 """Reconstruct an array view on a memory mapped file."""
239 if mode == 'w+':
240 # Do not zero the original data when unpickling
241 mode = 'r+'
243 if strides is None:
244 # Simple, contiguous memmap
245 return make_memmap(
246 filename, dtype=dtype, shape=shape, mode=mode, offset=offset,
247 order=order, unlink_on_gc_collect=unlink_on_gc_collect
248 )
249 else:
250 # For non-contiguous data, memmap the total enclosing buffer and then
251 # extract the non-contiguous view with the stride-tricks API
252 base = make_memmap(
253 filename, dtype=dtype, shape=total_buffer_len, offset=offset,
254 mode=mode, order=order, unlink_on_gc_collect=unlink_on_gc_collect
255 )
256 return as_strided(base, shape=shape, strides=strides)
259def _reduce_memmap_backed(a, m):
260 """Pickling reduction for memmap backed arrays.
262 a is expected to be an instance of np.ndarray (or np.memmap)
263 m is expected to be an instance of np.memmap on the top of the ``base``
264 attribute ancestry of a. ``m.base`` should be the real python mmap object.
265 """
266 # offset that comes from the striding differences between a and m
267 util.debug('[MEMMAP REDUCE] reducing a memmap-backed array '
268 '(shape, {}, pid: {})'.format(a.shape, os.getpid()))
269 a_start, a_end = np.byte_bounds(a)
270 m_start = np.byte_bounds(m)[0]
271 offset = a_start - m_start
273 # offset from the backing memmap
274 offset += m.offset
276 if m.flags['F_CONTIGUOUS']:
277 order = 'F'
278 else:
279 # The backing memmap buffer is necessarily contiguous hence C if not
280 # Fortran
281 order = 'C'
283 if a.flags['F_CONTIGUOUS'] or a.flags['C_CONTIGUOUS']:
284 # If the array is a contiguous view, no need to pass the strides
285 strides = None
286 total_buffer_len = None
287 else:
288 # Compute the total number of items to map from which the strided
289 # view will be extracted.
290 strides = a.strides
291 total_buffer_len = (a_end - a_start) // a.itemsize
293 return (_strided_from_memmap,
294 (m.filename, a.dtype, m.mode, offset, order, a.shape, strides,
295 total_buffer_len, False))
298def reduce_array_memmap_backward(a):
299 """reduce a np.array or a np.memmap from a child process"""
300 m = _get_backing_memmap(a)
301 if isinstance(m, np.memmap) and m.filename not in JOBLIB_MMAPS:
302 # if a is backed by a memmaped file, reconstruct a using the
303 # memmaped file.
304 return _reduce_memmap_backed(a, m)
305 else:
306 # a is either a regular (not memmap-backed) numpy array, or an array
307 # backed by a shared temporary file created by joblib. In the latter
308 # case, in order to limit the lifespan of these temporary files, we
309 # serialize the memmap as a regular numpy array, and decref the
310 # file backing the memmap (done implicitly in a previously registered
311 # finalizer, see ``unlink_on_gc_collect`` for more details)
312 return (
313 loads, (dumps(np.asarray(a), protocol=HIGHEST_PROTOCOL), )
314 )
317class ArrayMemmapForwardReducer(object):
318 """Reducer callable to dump large arrays to memmap files.
320 Parameters
321 ----------
322 max_nbytes: int
323 Threshold to trigger memmapping of large arrays to files created
324 a folder.
325 temp_folder_resolver: callable
326 An callable in charge of resolving a temporary folder name where files
327 for backing memmapped arrays are created.
328 mmap_mode: 'r', 'r+' or 'c'
329 Mode for the created memmap datastructure. See the documentation of
330 numpy.memmap for more details. Note: 'w+' is coerced to 'r+'
331 automatically to avoid zeroing the data on unpickling.
332 verbose: int, optional, 0 by default
333 If verbose > 0, memmap creations are logged.
334 If verbose > 1, both memmap creations, reuse and array pickling are
335 logged.
336 prewarm: bool, optional, False by default.
337 Force a read on newly memmapped array to make sure that OS pre-cache it
338 memory. This can be useful to avoid concurrent disk access when the
339 same data array is passed to different worker processes.
340 """
342 def __init__(self, max_nbytes, temp_folder_resolver, mmap_mode,
343 unlink_on_gc_collect, verbose=0, prewarm=True):
344 self._max_nbytes = max_nbytes
345 self._temp_folder_resolver = temp_folder_resolver
346 self._mmap_mode = mmap_mode
347 self.verbose = int(verbose)
348 if prewarm == "auto":
349 self._prewarm = not self._temp_folder.startswith(
350 SYSTEM_SHARED_MEM_FS
351 )
352 else:
353 self._prewarm = prewarm
354 self._prewarm = prewarm
355 self._memmaped_arrays = _WeakArrayKeyMap()
356 self._temporary_memmaped_filenames = set()
357 self._unlink_on_gc_collect = unlink_on_gc_collect
359 @property
360 def _temp_folder(self):
361 return self._temp_folder_resolver()
363 def __reduce__(self):
364 # The ArrayMemmapForwardReducer is passed to the children processes: it
365 # needs to be pickled but the _WeakArrayKeyMap need to be skipped as
366 # it's only guaranteed to be consistent with the parent process memory
367 # garbage collection.
368 # Although this reducer is pickled, it is not needed in its destination
369 # process (child processes), as we only use this reducer to send
370 # memmaps from the parent process to the children processes. For this
371 # reason, we can afford skipping the resolver, (which would otherwise
372 # be unpicklable), and pass it as None instead.
373 args = (self._max_nbytes, None, self._mmap_mode,
374 self._unlink_on_gc_collect)
375 kwargs = {
376 'verbose': self.verbose,
377 'prewarm': self._prewarm,
378 }
379 return ArrayMemmapForwardReducer, args, kwargs
381 def __call__(self, a):
382 m = _get_backing_memmap(a)
383 if m is not None and isinstance(m, np.memmap):
384 # a is already backed by a memmap file, let's reuse it directly
385 return _reduce_memmap_backed(a, m)
387 if (not a.dtype.hasobject and self._max_nbytes is not None and
388 a.nbytes > self._max_nbytes):
389 # check that the folder exists (lazily create the pool temp folder
390 # if required)
391 try:
392 os.makedirs(self._temp_folder)
393 os.chmod(self._temp_folder, FOLDER_PERMISSIONS)
394 except OSError as e:
395 if e.errno != errno.EEXIST:
396 raise e
398 try:
399 basename = self._memmaped_arrays.get(a)
400 except KeyError:
401 # Generate a new unique random filename. The process and thread
402 # ids are only useful for debugging purpose and to make it
403 # easier to cleanup orphaned files in case of hard process
404 # kill (e.g. by "kill -9" or segfault).
405 basename = "{}-{}-{}.pkl".format(
406 os.getpid(), id(threading.current_thread()), uuid4().hex)
407 self._memmaped_arrays.set(a, basename)
408 filename = os.path.join(self._temp_folder, basename)
410 # In case the same array with the same content is passed several
411 # times to the pool subprocess children, serialize it only once
413 is_new_memmap = filename not in self._temporary_memmaped_filenames
415 # add the memmap to the list of temporary memmaps created by joblib
416 self._temporary_memmaped_filenames.add(filename)
418 if self._unlink_on_gc_collect:
419 # Bump reference count of the memmap by 1 to account for
420 # shared usage of the memmap by a child process. The
421 # corresponding decref call will be executed upon calling
422 # resource_tracker.maybe_unlink, registered as a finalizer in
423 # the child.
424 # the incref/decref calls here are only possible when the child
425 # and the parent share the same resource_tracker. It is not the
426 # case for the multiprocessing backend, but it does not matter
427 # because unlinking a memmap from a child process is only
428 # useful to control the memory usage of long-lasting child
429 # processes, while the multiprocessing-based pools terminate
430 # their workers at the end of a map() call.
431 resource_tracker.register(filename, "file")
433 if is_new_memmap:
434 # Incref each temporary memmap created by joblib one extra
435 # time. This means that these memmaps will only be deleted
436 # once an extra maybe_unlink() is called, which is done once
437 # all the jobs have completed (or been canceled) in the
438 # Parallel._terminate_backend() method.
439 resource_tracker.register(filename, "file")
441 if not os.path.exists(filename):
442 util.debug(
443 "[ARRAY DUMP] Pickling new array (shape={}, dtype={}) "
444 "creating a new memmap at {}".format(
445 a.shape, a.dtype, filename))
446 for dumped_filename in dump(a, filename):
447 os.chmod(dumped_filename, FILE_PERMISSIONS)
449 if self._prewarm:
450 # Warm up the data by accessing it. This operation ensures
451 # that the disk access required to create the memmapping
452 # file are performed in the reducing process and avoids
453 # concurrent memmap creation in multiple children
454 # processes.
455 load(filename, mmap_mode=self._mmap_mode).max()
457 else:
458 util.debug(
459 "[ARRAY DUMP] Pickling known array (shape={}, dtype={}) "
460 "reusing memmap file: {}".format(
461 a.shape, a.dtype, os.path.basename(filename)))
463 # The worker process will use joblib.load to memmap the data
464 return (
465 (load_temporary_memmap, (filename, self._mmap_mode,
466 self._unlink_on_gc_collect))
467 )
468 else:
469 # do not convert a into memmap, let pickler do its usual copy with
470 # the default system pickler
471 util.debug(
472 '[ARRAY DUMP] Pickling array (NO MEMMAPPING) (shape={}, '
473 ' dtype={}).'.format(a.shape, a.dtype))
474 return (loads, (dumps(a, protocol=HIGHEST_PROTOCOL),))
477def get_memmapping_reducers(
478 forward_reducers=None, backward_reducers=None,
479 temp_folder_resolver=None, max_nbytes=1e6, mmap_mode='r', verbose=0,
480 prewarm=False, unlink_on_gc_collect=True, **kwargs):
481 """Construct a pair of memmapping reducer linked to a tmpdir.
483 This function manage the creation and the clean up of the temporary folders
484 underlying the memory maps and should be use to get the reducers necessary
485 to construct joblib pool or executor.
486 """
487 if forward_reducers is None:
488 forward_reducers = dict()
489 if backward_reducers is None:
490 backward_reducers = dict()
492 if np is not None:
493 # Register smart numpy.ndarray reducers that detects memmap backed
494 # arrays and that is also able to dump to memmap large in-memory
495 # arrays over the max_nbytes threshold
496 forward_reduce_ndarray = ArrayMemmapForwardReducer(
497 max_nbytes, temp_folder_resolver, mmap_mode, unlink_on_gc_collect,
498 verbose, prewarm=prewarm)
499 forward_reducers[np.ndarray] = forward_reduce_ndarray
500 forward_reducers[np.memmap] = forward_reduce_ndarray
502 # Communication from child process to the parent process always
503 # pickles in-memory numpy.ndarray without dumping them as memmap
504 # to avoid confusing the caller and make it tricky to collect the
505 # temporary folder
506 backward_reducers[np.ndarray] = reduce_array_memmap_backward
507 backward_reducers[np.memmap] = reduce_array_memmap_backward
509 return forward_reducers, backward_reducers
512class TemporaryResourcesManager(object):
513 """Stateful object able to manage temporary folder and pickles
515 It exposes:
516 - a per-context folder name resolving API that memmap-based reducers will
517 rely on to know where to pickle the temporary memmaps
518 - a temporary file/folder management API that internally uses the
519 resource_tracker.
520 """
522 def __init__(self, temp_folder_root=None, context_id=None):
523 self._current_temp_folder = None
524 self._temp_folder_root = temp_folder_root
525 self._use_shared_mem = None
526 self._cached_temp_folders = dict()
527 self._id = uuid4().hex
528 self._finalizers = {}
529 if context_id is None:
530 # It would be safer to not assign a default context id (less silent
531 # bugs), but doing this while maintaining backward compatibility
532 # with the previous, context-unaware version get_memmaping_executor
533 # exposes too many low-level details.
534 context_id = uuid4().hex
535 self.set_current_context(context_id)
537 def set_current_context(self, context_id):
538 self._current_context_id = context_id
539 self.register_new_context(context_id)
541 def register_new_context(self, context_id):
542 # Prepare a sub-folder name specific to a context (usually a unique id
543 # generated by each instance of the Parallel class). Do not create in
544 # advance to spare FS write access if no array is to be dumped).
545 if context_id in self._cached_temp_folders:
546 return
547 else:
548 # During its lifecycle, one Parallel object can have several
549 # executors associated to it (for instance, if a loky worker raises
550 # an exception, joblib shutdowns the executor and instantly
551 # recreates a new one before raising the error - see
552 # ``ensure_ready``. Because we don't want two executors tied to
553 # the same Parallel object (and thus the same context id) to
554 # register/use/delete the same folder, we also add an id specific
555 # to the current Manager (and thus specific to its associated
556 # executor) to the folder name.
557 new_folder_name = (
558 "joblib_memmapping_folder_{}_{}_{}".format(
559 os.getpid(), self._id, context_id)
560 )
561 new_folder_path, _ = _get_temp_dir(
562 new_folder_name, self._temp_folder_root
563 )
564 self.register_folder_finalizer(new_folder_path, context_id)
565 self._cached_temp_folders[context_id] = new_folder_path
567 def resolve_temp_folder_name(self):
568 """Return a folder name specific to the currently activated context"""
569 return self._cached_temp_folders[self._current_context_id]
571 # resource management API
573 def register_folder_finalizer(self, pool_subfolder, context_id):
574 # Register the garbage collector at program exit in case caller forgets
575 # to call terminate explicitly: note we do not pass any reference to
576 # ensure that this callback won't prevent garbage collection of
577 # parallel instance and related file handler resources such as POSIX
578 # semaphores and pipes
579 pool_module_name = whichmodule(delete_folder, 'delete_folder')
580 resource_tracker.register(pool_subfolder, "folder")
582 def _cleanup():
583 # In some cases the Python runtime seems to set delete_folder to
584 # None just before exiting when accessing the delete_folder
585 # function from the closure namespace. So instead we reimport
586 # the delete_folder function explicitly.
587 # https://github.com/joblib/joblib/issues/328
588 # We cannot just use from 'joblib.pool import delete_folder'
589 # because joblib should only use relative imports to allow
590 # easy vendoring.
591 delete_folder = __import__(
592 pool_module_name, fromlist=['delete_folder']
593 ).delete_folder
594 try:
595 delete_folder(pool_subfolder, allow_non_empty=True)
596 resource_tracker.unregister(pool_subfolder, "folder")
597 except OSError:
598 warnings.warn("Failed to delete temporary folder: {}"
599 .format(pool_subfolder))
601 self._finalizers[context_id] = atexit.register(_cleanup)
603 def _clean_temporary_resources(self, context_id=None, force=False,
604 allow_non_empty=False):
605 """Clean temporary resources created by a process-based pool"""
606 if context_id is None:
607 # Iterates over a copy of the cache keys to avoid Error due to
608 # iterating over a changing size dictionary.
609 for context_id in list(self._cached_temp_folders):
610 self._clean_temporary_resources(
611 context_id, force=force, allow_non_empty=allow_non_empty
612 )
613 else:
614 temp_folder = self._cached_temp_folders.get(context_id)
615 if temp_folder and os.path.exists(temp_folder):
616 for filename in os.listdir(temp_folder):
617 if force:
618 # Some workers have failed and the ref counted might
619 # be off. The workers should have shut down by this
620 # time so forcefully clean up the files.
621 resource_tracker.unregister(
622 os.path.join(temp_folder, filename), "file"
623 )
624 else:
625 resource_tracker.maybe_unlink(
626 os.path.join(temp_folder, filename), "file"
627 )
629 # When forcing clean-up, try to delete the folder even if some
630 # files are still in it. Otherwise, try to delete the folder
631 allow_non_empty |= force
633 # Clean up the folder if possible, either if it is empty or
634 # if none of the files in it are in used and allow_non_empty.
635 try:
636 delete_folder(
637 temp_folder, allow_non_empty=allow_non_empty
638 )
639 # Forget the folder once it has been deleted
640 self._cached_temp_folders.pop(context_id, None)
641 resource_tracker.unregister(temp_folder, "folder")
643 # Also cancel the finalizers that gets triggered at gc.
644 finalizer = self._finalizers.pop(context_id, None)
645 if finalizer is not None:
646 atexit.unregister(finalizer)
648 except OSError:
649 # Temporary folder cannot be deleted right now.
650 # This folder will be cleaned up by an atexit
651 # finalizer registered by the memmapping_reducer.
652 pass