Transactions¶
Tortoise ORM provides a simple way to manage transactions. You can use the
atomic() decorator or in_transaction() context manager.
atomic() and in_transaction() can be nested. The inner blocks will create transaction savepoints,
and if an exception is raised and then caught outside of a nested block, the transaction will be rolled back
to the state before the block was entered. The outermost block will be the one that actually commits the transaction.
The savepoints are supported for Postgres, MySQL, MSSQL and SQLite. For other databases, it is advised to
propagate the exception to the outermost block to ensure that the transaction is rolled back.
# this block will commit changes on exit async with in_transaction(): await MyModel.create(name='foo') try: # this block will create a savepoint and rollback to it if an exception is raised async with in_transaction(): await MyModel.create(name='bar') # this will rollback to the savepoint, meaning that # the 'bar' record will not be created, however, # the 'foo' record will be created raise Exception() except Exception: pass
When using asyncio.gather or similar ways to spin up concurrent tasks in a transaction block,
avoid having nested transaction blocks in the concurrent tasks. Transactions are stateful and nested
blocks are expected to run sequentially, not concurrently.
-
tortoise.transactions.atomic(connection_name=
None)[source]¶ Transaction decorator.
You can wrap your function with this decorator to run it into one transaction. If error occurs transaction will rollback.
-
tortoise.transactions.in_transaction(connection_name=
None)[source]¶ Transaction context manager.
You can run your code inside
async with in_transaction():statement to run it into one transaction. If error occurs transaction will rollback.