Having a storage, we then use it to instantiate a database, which we
then connect to by calling ``open()``. A process with multiple
threads will often have multiple connections to the same database,
with different threads having different connections.
There are other installation mechanisms available to manage the
installation of Python packages. This tutorial assumes that you are using
a plain Python installation and install ZODB globally.
There are a number of convenient shortcuts you can use for some of the
commonly used storages:
- You can pass a file name to the ``DB`` constructor to have it construct
a FileStorage for you::
Configuration
=============
db = ZODB.DB('mydata.fs')
When a program wants to use the ZODB it has to establish a connection, like any other database. For the ZODB we need 3 different parts: a storage, a database and finally a connection:
You can pass None to create an in-memory database::
>>> from ZODB.FileStorage import FileStorage
>>> from ZODB.DB import DB
>>> storage = FileStorage('Data.fs')
>>> db = DB(storage)
>>> connection = db.open()
>>> root = connection.root()
memory_db = ZODB.DB(None)
We create a storage called FileStorage, which is the current standard storage
used by virtually everyone. It keeps track of all data in a single file as
stated by the first parameter. From this storage we create a database and
finally open up a connection. Finally, we retrieve the database root object
from the connection that we opened.
- If you're only going to use one connection, you can call the
``connection`` function::
connection = ZODB.connection('mydata.fs')
memory_connection = ZODB.connection(None)
Storing objects
===============
To store an object in the ZODB we simply attach it to any other object that
already lives in the database. Hence, the root object functions as a
boot-strapping point. The root object is a dictionary and you can start storing
your objects directly in there:
To store an object in the ZODB we simply attach it to any other object
that already lives in the database. Hence, the root object functions
as a boot-strapping point. The root object is meant to serve as a
namespace for top-level objects in your database. We could store
account objects directly on the root object::
import account
# Probably a bad idea:
root.account1 = account.Account()
But if you're going to store many objects, you'll want to use a
collection object [#root]_::
import account, BTrees.OOBTree
root.accounts = BTrees.OOBTree.BTree()
root.accounts['account-1'] = Account()
Another common practice is to store a persistent object in the root of
the database that provides an application-specific root::
root.accounts = AccountManagementApplication()
That can facilitate encapsulation of an application that shares a
database with other applications. This is a little bit like using
modules to avoid namespace colisions in Python programs.
>>> root['account-1'] = Account()
>>> root['account-2'] = Account()
Containers and search
=====================
Frameworks like Zope only
create a single object in the ZODB root representing the application itself and
then let all sub-objects be referenced from there on. They choose names like
'app' for the first object they place in the ZODB.
BTrees provide the core scalable containers and indexing facility for
ZODB. There are different families of BTrees. The most general are
OOBTrees, which have object keys and values. There are specialized
BTrees that support integer keys and values. Integers can be stored
more efficiently, and compared more quickly than objects and their
often used as application-level object identifiers. It's critical,
when using BTrees, to make sure that it's keys have a stable ordering.
ZODB doesn't provide a query engine. The primary way to access
objects in ZODB is by traversing (accessing attributes or items, or
calling methods) other objects. Object traversal is typically much
faster than search.
You can use BTrees to build indexes for efficient search, when
necessary. If your application is search centric, or if you prefer to
approach data access that way, then ZODB might not be the best
technology for you.
Transactions
============
You now have two objects placed in your root object and in your database.
However, they are not permanently stored yet. The ZODB uses transactions and to
make your changes permanent, you have to commit the transaction:
You now have objects in your root object and in your database.
However, they are not permanently stored yet. The ZODB uses
transactions and to make your changes permanent, you have to commit
the transaction::
>>> import transaction
>>> transaction.commit()
>>> root.keys()
['account-1', 'account-2']
import transaction
Now you can stop and start your application and look at the root object again,
and you will find the entries 'account-1' and 'account-2' to be still be
present and be the objects you created.
transaction.commit()
Objects that have not been stored in
the ZODB yet are not rolled back by an abort.
Now you can stop and start your application and look at the root object again,
and you will find the data you saved.
If your application makes changes during a transaction and finds that it does
not want to commit those changes, then you can abort the transaction and have
the changes rolled back for you:
the changes rolled back [#rollback]_ for you::
>>> del root['account-1']
>>> root.keys()
['account-2']
>>> transaction.abort()
>>> root.keys()
['account-1', 'account-2']
Persistent objects
==================
transaction.abort()
One last aspect that we need to cover are persistent objects themselves. The
ZODB will be happy to store almost any Python object that you pass to it (it
won't store files for example). But in order for noticing which objects have
changed the ZODB needs those objects to cooperate with the database. In
general, you just subclass `persistent.Persistent` to make this happen. So our
example class from above would read like::
import persistent
class Account(persistent.Persistent):
# ... same code as above ...
Transactions are a very powerful way to protect the integrity of a
database. Transactions have the property that all of the changes made
in a transaction are saved, or none of them are. If in the midst of a
program, there's an error after making changes, you can simply abort
the transaction (or not commit it) and all of the intermediate changes
you make are automatically discarded.
..
Memory Management
=================
Have a look at the reference documentation about the rules of persistency and
to find out more about specialised persistent objects like BTrees.
ZODB manages moving objects in and out of memory for you. The unit of
storage is the persistent object. When you access attributes of a
persistent objects, it's loaded from the database automatically, if
necessary. If too many objects are in memory, then objects used least
recently are evicted [#eviction]_. The maximum number of objects or
bytes in memory is configurable.
Summary
=======
You have seen how to install ZODB and how to open a database in your
application and to start storing objects in it. We also touched the two simple
transaction commands: commit and abort. The reference documentation contains
sections with more information on the individual topics.
application and to start storing objects in it. We also touched the
two simple transaction commands: ``commit`` and ``abort``. The
reference documentation contains sections with more information on the
individual topics.
.. [#changed]
You can manually mark an object as changed by setting it's
``_p_changed__`` attribute to ``True``. You might do this if you
update a subobject, such as a standard Python ``list`` or ``set``,
that doesn't subclass ``Persistent``.
.. [#persistentclasses]
Actually, there is semi-experimental support for storing classes in
the database, but applications rarely do this.
.. [#root]
The root object is a fairy simple persistent object that's stored
in a single database record. If you stored many objects in it,
it's database record would become very large, causing updates to be
inefficient and causing memory to be used ineffeciently.
Another reason not to store items directly in the root object is
that doing so would make adding a second collection of objects
later awkward.
.. [#rollback]
A caveat is that ZODB can only roll back changes to objects that
have been stored and committed to the database. Objects not
previously committed can't be rolled back because there's no
previous state to roll back to.
.. [#eviction]
Objects aren't actually evicted, but their state is released, so
they take up much less memory and any objects they referenced can
