psycopg2 integration with transaction
This package provides two-phase commit integration between psycopg2 and the transaction package.
>>> import psycopg2, psycopg2transaction, transaction >>> conn = psycopg2.connect(mydsn) >>> trans = transaction.get() >>> psycopg2transaction.join(conn, trans) >>> ... do things with the psycopg2 connection >>> transaction.commit()
This is just an example. The connection might come from a pool or otherwise already exist, as long as it is in a suitable state to start a distributed transaction.
Similarly, the transaction may come from the thread-local transaction manager, as shown above, or from an application-managed transaction manager.
The critical line in the example above is:
>>> psycopg2transaction.join(conn, trans)
The work done in the connection is committed by calling the
transaction's commit method, typically by calling
commit on the
transaction's manager, as shown above.
If you use a thread-local transaction manager (as in the example
above), you can omit the transaction argumet to
Limited connection management is provided:
>>> dsn = 'dbname=test' >>> conn = psycopg2transaction.join(dsn) ... >>> conn = psycopg2transaction.join(dsn) ... >>> transaction.commit()
If you pass a connection string rather than a connection to
connection will be opened for you and closed when the transaction is
join is called multiple times with the same connection
string, then the same connection is used and returned.
Using Postgres notify
You can't use two-phase commit in postgres transactions in which
you've used NOTIFY (or LISTEN or UNLISTEN). You can cause NOTIFY to
be used after a transaction has committed, but before a managed
connection has been closed by passing a
notify argument to join
with a name of a notification or a sequence of string notification
named. The strings may be simple names, or names followed by a comma
and string data, as would follow NOTIFY in a notify expression:
>>> conn = psycopg2transaction.join(dsn, notify='myjobs')