|Version 8 (modified by 9 years ago) (diff),|
Part of DjangoSpecifications
Threading improvements in Django
According to tickets #5632, #6950 and discussions http://groups.google.com/group/django-users/browse_frm/thread/a7d42475b66530bd, http://groups.google.com/group/django-developers/browse_thread/thread/fbcfa88c997d1bb3, at least the following components of Django are not entirely thread-safe:
This specification outlines the changes that need to be implemented to solve these issues.
A related task is to identify other components not mentioned here that have threading issues.
See #1442 and http://groups.google.com/group/django-developers/browse_thread/thread/905f79e350525c95 for threading discussions.
Instead of ad-hoc solutions like the one proposed in #5632, there should be a locking module that can be reused.
Five types of locking primitives could be implemented:
- Django-wide global locks
- module-level locks
- class-level locks
- instance-level locks
- function-level locks
I don't see any use for a Django-wide lock. Module-level lock is needed for sessions. Class-level locks seem appropriate for models. Template loader can use a function-level lock.
FIXME: module and class level locks missing.
The implementation is a modified copy of http://www.phyast.pitt.edu/~micheles/python/documentation.html and should perhaps reside in
from django.utils._threading_local import RLock def getattr_(obj, name, default_thunk): "Similar to .setdefault in dictionaries." try: return getattr(obj, name) except AttributeError: default = default_thunk() setattr(obj, name, default) return default def locked(func, *args, **kw): lock = getattr_(func, "__lock", RLock) lock.acquire() try: result = func(*args, **kw) finally: lock.release() return result
Database model locking
All functions that have
.alters_data = True should be wrapped with a locked block.
Once older python version support will be dropped in distant future, locking should be implemented with
Although this sounds like a horrible hack, perhaps a portable wrapper around
mmap.mmap can be used for emulating shared memory.