django1/django/db/transaction.py

536 lines
18 KiB
Python

"""
This module implements a transaction manager that can be used to define
transaction handling in a request or view function. It is used by transaction
control middleware and decorators.
The transaction manager can be in managed or in auto state. Auto state means the
system is using a commit-on-save strategy (actually it's more like
commit-on-change). As soon as the .save() or .delete() (or related) methods are
called, a commit is made.
Managed transactions don't do those commits, but will need some kind of manual
or implicit commits or rollbacks.
"""
import warnings
from functools import wraps
from django.db import (
connections, DEFAULT_DB_ALIAS,
DatabaseError, ProgrammingError)
from django.utils.decorators import available_attrs
class TransactionManagementError(ProgrammingError):
"""
This exception is thrown when transaction management is used improperly.
"""
pass
################
# Private APIs #
################
def get_connection(using=None):
"""
Get a database connection by name, or the default database connection
if no name is provided.
"""
if using is None:
using = DEFAULT_DB_ALIAS
return connections[using]
###########################
# Deprecated private APIs #
###########################
def abort(using=None):
"""
Roll back any ongoing transactions and clean the transaction management
state of the connection.
This method is to be used only in cases where using balanced
leave_transaction_management() calls isn't possible. For example after a
request has finished, the transaction state isn't known, yet the connection
must be cleaned up for the next request.
"""
get_connection(using).abort()
def enter_transaction_management(managed=True, using=None, forced=False):
"""
Enters transaction management for a running thread. It must be balanced with
the appropriate leave_transaction_management call, since the actual state is
managed as a stack.
The state and dirty flag are carried over from the surrounding block or
from the settings, if there is no surrounding block (dirty is always false
when no current block is running).
"""
get_connection(using).enter_transaction_management(managed, forced)
def leave_transaction_management(using=None):
"""
Leaves transaction management for a running thread. A dirty flag is carried
over to the surrounding block, as a commit will commit all changes, even
those from outside. (Commits are on connection level.)
"""
get_connection(using).leave_transaction_management()
def is_dirty(using=None):
"""
Returns True if the current transaction requires a commit for changes to
happen.
"""
return get_connection(using).is_dirty()
def set_dirty(using=None):
"""
Sets a dirty flag for the current thread and code streak. This can be used
to decide in a managed block of code to decide whether there are open
changes waiting for commit.
"""
get_connection(using).set_dirty()
def set_clean(using=None):
"""
Resets a dirty flag for the current thread and code streak. This can be used
to decide in a managed block of code to decide whether a commit or rollback
should happen.
"""
get_connection(using).set_clean()
def is_managed(using=None):
warnings.warn("'is_managed' is deprecated.",
DeprecationWarning, stacklevel=2)
def managed(flag=True, using=None):
warnings.warn("'managed' no longer serves a purpose.",
DeprecationWarning, stacklevel=2)
def commit_unless_managed(using=None):
warnings.warn("'commit_unless_managed' is now a no-op.",
DeprecationWarning, stacklevel=2)
def rollback_unless_managed(using=None):
warnings.warn("'rollback_unless_managed' is now a no-op.",
DeprecationWarning, stacklevel=2)
###############
# Public APIs #
###############
def get_autocommit(using=None):
"""
Get the autocommit status of the connection.
"""
return get_connection(using).get_autocommit()
def set_autocommit(autocommit, using=None):
"""
Set the autocommit status of the connection.
"""
return get_connection(using).set_autocommit(autocommit)
def commit(using=None):
"""
Commits a transaction and resets the dirty flag.
"""
get_connection(using).commit()
def rollback(using=None):
"""
Rolls back a transaction and resets the dirty flag.
"""
get_connection(using).rollback()
def savepoint(using=None):
"""
Creates a savepoint (if supported and required by the backend) inside the
current transaction. Returns an identifier for the savepoint that will be
used for the subsequent rollback or commit.
"""
return get_connection(using).savepoint()
def savepoint_rollback(sid, using=None):
"""
Rolls back the most recent savepoint (if one exists). Does nothing if
savepoints are not supported.
"""
get_connection(using).savepoint_rollback(sid)
def savepoint_commit(sid, using=None):
"""
Commits the most recent savepoint (if one exists). Does nothing if
savepoints are not supported.
"""
get_connection(using).savepoint_commit(sid)
def clean_savepoints(using=None):
"""
Resets the counter used to generate unique savepoint ids in this thread.
"""
get_connection(using).clean_savepoints()
def get_rollback(using=None):
"""
Gets the "needs rollback" flag -- for *advanced use* only.
"""
return get_connection(using).get_rollback()
def set_rollback(rollback, using=None):
"""
Sets or unsets the "needs rollback" flag -- for *advanced use* only.
When `rollback` is `True`, it triggers a rollback when exiting the
innermost enclosing atomic block that has `savepoint=True` (that's the
default). Use this to force a rollback without raising an exception.
When `rollback` is `False`, it prevents such a rollback. Use this only
after rolling back to a known-good state! Otherwise, you break the atomic
block and data corruption may occur.
"""
return get_connection(using).set_rollback(rollback)
#################################
# Decorators / context managers #
#################################
class Atomic(object):
"""
This class guarantees the atomic execution of a given block.
An instance can be used either as a decorator or as a context manager.
When it's used as a decorator, __call__ wraps the execution of the
decorated function in the instance itself, used as a context manager.
When it's used as a context manager, __enter__ creates a transaction or a
savepoint, depending on whether a transaction is already in progress, and
__exit__ commits the transaction or releases the savepoint on normal exit,
and rolls back the transaction or to the savepoint on exceptions.
It's possible to disable the creation of savepoints if the goal is to
ensure that some code runs within a transaction without creating overhead.
A stack of savepoints identifiers is maintained as an attribute of the
connection. None denotes the absence of a savepoint.
This allows reentrancy even if the same AtomicWrapper is reused. For
example, it's possible to define `oa = @atomic('other')` and use `@oa` or
`with oa:` multiple times.
Since database connections are thread-local, this is thread-safe.
"""
def __init__(self, using, savepoint):
self.using = using
self.savepoint = savepoint
def __enter__(self):
connection = get_connection(self.using)
if not connection.in_atomic_block:
# Reset state when entering an outermost atomic block.
connection.commit_on_exit = True
connection.needs_rollback = False
if not connection.get_autocommit():
# Some database adapters (namely sqlite3) don't handle
# transactions and savepoints properly when autocommit is off.
# Turning autocommit back on isn't an option; it would trigger
# a premature commit. Give up if that happens.
if connection.features.autocommits_when_autocommit_is_off:
raise TransactionManagementError(
"Your database backend doesn't behave properly when "
"autocommit is off. Turn it on before using 'atomic'.")
# When entering an atomic block with autocommit turned off,
# Django should only use savepoints and shouldn't commit.
# This requires at least a savepoint for the outermost block.
if not self.savepoint:
raise TransactionManagementError(
"The outermost 'atomic' block cannot use "
"savepoint = False when autocommit is off.")
# Pretend we're already in an atomic block to bypass the code
# that disables autocommit to enter a transaction, and make a
# note to deal with this case in __exit__.
connection.in_atomic_block = True
connection.commit_on_exit = False
if connection.in_atomic_block:
# We're already in a transaction; create a savepoint, unless we
# were told not to or we're already waiting for a rollback. The
# second condition avoids creating useless savepoints and prevents
# overwriting needs_rollback until the rollback is performed.
if self.savepoint and not connection.needs_rollback:
sid = connection.savepoint()
connection.savepoint_ids.append(sid)
else:
connection.savepoint_ids.append(None)
else:
# We aren't in a transaction yet; create one.
# The usual way to start a transaction is to turn autocommit off.
# However, some database adapters (namely sqlite3) don't handle
# transactions and savepoints properly when autocommit is off.
# In such cases, start an explicit transaction instead, which has
# the side-effect of disabling autocommit.
if connection.features.autocommits_when_autocommit_is_off:
connection._start_transaction_under_autocommit()
connection.autocommit = False
else:
connection.set_autocommit(False)
connection.in_atomic_block = True
def __exit__(self, exc_type, exc_value, traceback):
connection = get_connection(self.using)
if connection.savepoint_ids:
sid = connection.savepoint_ids.pop()
else:
# Prematurely unset this flag to allow using commit or rollback.
connection.in_atomic_block = False
try:
if exc_type is None and not connection.needs_rollback:
if connection.in_atomic_block:
# Release savepoint if there is one
if sid is not None:
try:
connection.savepoint_commit(sid)
except DatabaseError:
connection.savepoint_rollback(sid)
raise
else:
# Commit transaction
try:
connection.commit()
except DatabaseError:
connection.rollback()
raise
else:
# This flag will be set to True again if there isn't a savepoint
# allowing to perform the rollback at this level.
connection.needs_rollback = False
if connection.in_atomic_block:
# Roll back to savepoint if there is one, mark for rollback
# otherwise.
if sid is None:
connection.needs_rollback = True
else:
connection.savepoint_rollback(sid)
else:
# Roll back transaction
connection.rollback()
finally:
# Outermost block exit when autocommit was enabled.
if not connection.in_atomic_block:
if connection.features.autocommits_when_autocommit_is_off:
connection.autocommit = True
else:
connection.set_autocommit(True)
# Outermost block exit when autocommit was disabled.
elif not connection.savepoint_ids and not connection.commit_on_exit:
connection.in_atomic_block = False
def __call__(self, func):
@wraps(func, assigned=available_attrs(func))
def inner(*args, **kwargs):
with self:
return func(*args, **kwargs)
return inner
def atomic(using=None, savepoint=True):
# Bare decorator: @atomic -- although the first argument is called
# `using`, it's actually the function being decorated.
if callable(using):
return Atomic(DEFAULT_DB_ALIAS, savepoint)(using)
# Decorator: @atomic(...) or context manager: with atomic(...): ...
else:
return Atomic(using, savepoint)
def _non_atomic_requests(view, using):
try:
view._non_atomic_requests.add(using)
except AttributeError:
view._non_atomic_requests = set([using])
return view
def non_atomic_requests(using=None):
if callable(using):
return _non_atomic_requests(using, DEFAULT_DB_ALIAS)
else:
if using is None:
using = DEFAULT_DB_ALIAS
return lambda view: _non_atomic_requests(view, using)
############################################
# Deprecated decorators / context managers #
############################################
class Transaction(object):
"""
Acts as either a decorator, or a context manager. If it's a decorator it
takes a function and returns a wrapped function. If it's a contextmanager
it's used with the ``with`` statement. In either event entering/exiting
are called before and after, respectively, the function/block is executed.
autocommit, commit_on_success, and commit_manually contain the
implementations of entering and exiting.
"""
def __init__(self, entering, exiting, using):
self.entering = entering
self.exiting = exiting
self.using = using
def __enter__(self):
self.entering(self.using)
def __exit__(self, exc_type, exc_value, traceback):
self.exiting(exc_type, self.using)
def __call__(self, func):
@wraps(func)
def inner(*args, **kwargs):
with self:
return func(*args, **kwargs)
return inner
def _transaction_func(entering, exiting, using):
"""
Takes 3 things, an entering function (what to do to start this block of
transaction management), an exiting function (what to do to end it, on both
success and failure, and using which can be: None, indicating using is
DEFAULT_DB_ALIAS, a callable, indicating that using is DEFAULT_DB_ALIAS and
to return the function already wrapped.
Returns either a Transaction objects, which is both a decorator and a
context manager, or a wrapped function, if using is a callable.
"""
# Note that although the first argument is *called* `using`, it
# may actually be a function; @autocommit and @autocommit('foo')
# are both allowed forms.
if using is None:
using = DEFAULT_DB_ALIAS
if callable(using):
return Transaction(entering, exiting, DEFAULT_DB_ALIAS)(using)
return Transaction(entering, exiting, using)
def autocommit(using=None):
"""
Decorator that activates commit on save. This is Django's default behavior;
this decorator is useful if you globally activated transaction management in
your settings file and want the default behavior in some view functions.
"""
warnings.warn("autocommit is deprecated in favor of set_autocommit.",
DeprecationWarning, stacklevel=2)
def entering(using):
enter_transaction_management(managed=False, using=using)
def exiting(exc_type, using):
leave_transaction_management(using=using)
return _transaction_func(entering, exiting, using)
def commit_on_success(using=None):
"""
This decorator activates commit on response. This way, if the view function
runs successfully, a commit is made; if the viewfunc produces an exception,
a rollback is made. This is one of the most common ways to do transaction
control in Web apps.
"""
warnings.warn("commit_on_success is deprecated in favor of atomic.",
DeprecationWarning, stacklevel=2)
def entering(using):
enter_transaction_management(using=using)
def exiting(exc_type, using):
try:
if exc_type is not None:
if is_dirty(using=using):
rollback(using=using)
else:
if is_dirty(using=using):
try:
commit(using=using)
except:
rollback(using=using)
raise
finally:
leave_transaction_management(using=using)
return _transaction_func(entering, exiting, using)
def commit_manually(using=None):
"""
Decorator that activates manual transaction control. It just disables
automatic transaction control and doesn't do any commit/rollback of its
own -- it's up to the user to call the commit and rollback functions
themselves.
"""
warnings.warn("commit_manually is deprecated in favor of set_autocommit.",
DeprecationWarning, stacklevel=2)
def entering(using):
enter_transaction_management(using=using)
def exiting(exc_type, using):
leave_transaction_management(using=using)
return _transaction_func(entering, exiting, using)
def commit_on_success_unless_managed(using=None, savepoint=False):
"""
Transitory API to preserve backwards-compatibility while refactoring.
Once the legacy transaction management is fully deprecated, this should
simply be replaced by atomic. Until then, it's necessary to guarantee that
a commit occurs on exit, which atomic doesn't do when it's nested.
Unlike atomic, savepoint defaults to False because that's closer to the
legacy behavior.
"""
connection = get_connection(using)
if connection.get_autocommit() or connection.in_atomic_block:
return atomic(using, savepoint)
else:
def entering(using):
pass
def exiting(exc_type, using):
set_dirty(using=using)
return _transaction_func(entering, exiting, using)