fix(utility): add lock for cached file between processes #1151
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Pull Request Test Coverage Report for Build 1627169800
💛 - Coveralls |
linjiX
requested changes
Dec 6, 2021
tensorbay/utility/common.py
Outdated
| def wrapper(self: Any, *arg: Any, **kwargs: Any) -> None: | ||
| key = getattr(self, attr_name) if attr_name else id(self) | ||
| # https://github.com/PyCQA/pylint/issues/3313 | ||
| lock = process_locks.setdefault(key, manager.Lock()) # pylint: disable=no-member |
There was a problem hiding this comment.
get item from a Manager.dict() seems a copy behavior, which means the gotten lock is a copy of the lock in the dict. That means the operation in the outside lock will not affect the inside lock?
There was a problem hiding this comment.
Shared objects are capable of being nested. The manager.Lock() is also a shared object, thus the change will be synced to the dict.
https://docs.python.org/3.8/library/multiprocessing.html#multiprocessing.managers.SyncManager
Lee-000
force-pushed
the
T19988_add_multiprocess_lock_for_cache
branch
from
3d38af0 to
2490402
Compare
Lee-000
force-pushed
the
T19988_add_multiprocess_lock_for_cache
branch
from
2490402 to
b50bae0
Compare
linjiX
requested changes
Dec 7, 2021
Lee-000
force-pushed
the
T19988_add_multiprocess_lock_for_cache
branch
2 times, most recently
from
1858cae to
3a07341
Compare
AChenQ
approved these changes
Dec 8, 2021
Lee-000
force-pushed
the
T19988_add_multiprocess_lock_for_cache
branch
2 times, most recently
from
1a95b3f to
ba54946
Compare
linjiX
requested changes
Dec 10, 2021
tensorbay/utility/common.py
Outdated
| def wrapper(func_self: Any, *arg: Any, **kwargs: Any) -> None: | ||
| key = getattr(func_self, self._attr_name) | ||
| lock_exists = False | ||
| if key in self.process_locks: |
There was a problem hiding this comment.
will it possible that two processes go into this if statement at the same time, and two processes both return false, which makes them both create their own locks?
Lee-000
force-pushed
the
T19988_add_multiprocess_lock_for_cache
branch
from
ba54946 to
8fb8259
Compare
linjiX
requested changes
Dec 14, 2021
tensorbay/utility/common.py
Outdated
| # https://github.com/PyCQA/pylint/issues/3313 | ||
| self.process_locks[key] = self._manager.Lock() # pylint: disable=no-member | ||
| with _acquire(self.process_locks[key]) as success: | ||
| self._lock.release() |
There was a problem hiding this comment.
if there are something wrong happens before releasing the self._lock, will it cause self._lock never release?
linjiX
requested changes
Dec 14, 2021
| self._lock.release() | ||
| if success and not lock_exists: | ||
| func(func_self, *arg, **kwargs) | ||
| del self.process_locks[key] |
There was a problem hiding this comment.
It that possible one process is deleting the lock from self.process_locks, and at the same time another process is checking whether the key in the self.process_locks?
Lee-000
force-pushed
the
T19988_add_multiprocess_lock_for_cache
branch
from
8fb8259 to
12a54d7
Compare
linjiX
requested changes
Dec 15, 2021
Lee-000
force-pushed
the
T19988_add_multiprocess_lock_for_cache
branch
from
12a54d7 to
a7b26ef
Compare
Lee-000
force-pushed
the
T19988_add_multiprocess_lock_for_cache
branch
from
a7b26ef to
7797a2d
Compare
linjiX
approved these changes
Dec 28, 2021
When using PyTorch dataloader with multiple workers, the cache file will be accessed by different processes. And the cache file can be written and read at the same time, which would lead to the read content being empty. Thus, use a temp filename while writing the cache file and then change it to the cache path. And add a lock to avoid repeatedly writing cache. PR Closed: Graviti-AI#1151
Lee-000
force-pushed
the
T19988_add_multiprocess_lock_for_cache
branch
from
7797a2d to
abb7b9c
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
When using PyTorch dataloader with multiple workers, the cache file will
be accessed by different processes. And the cache file can be written
and read at the same time, which would lead to the read content being
empty.
Thus, use a temp filename while writing the cache file and then change
it to the cache path. And add a lock to avoid repeatedly writing cache.