Models¶
Models define a Python-ic interface to relational databases using the SQLAlchemy toolkit that provides a full suite of well known enterprise-level persistence patterns, designed for efficient and high-performing database access, adapted into a simple and Pythonic domain language [SQLAlchemy2013].
In order to add SQLAlchemy support to our application, the Flask-SQLAlchemy extension is used. It provides useful defaults as well as extra declarative base helpers. We recommend reading Official Tutorial for a full introduction and Other Tutorial for better understanding of ORM concepts.
Code structure¶
Our custom bridge contains several custom types and driver hacks for smoother integration with multiple database engines. The code structure follows:
invenio/ext/sqlalchemy
/engines
mysql.py
__init__.py
expressions.py
types.py
utils.py
Before you start writing a new model please take a look at the
db
object. It will also make it easier to
understand the following example of a simple model written using SQLAlchemy:
# General imports.
from invenio.ext.sqlalchemy import db
# Create your models here.
class User(db.Model):
"""Represents a User record."""
__tablename__ = 'user'
id = db.Column(db.Integer(15, unsigned=True), primary_key=True,
autoincrement=True)
email = db.Column(db.String(255), nullable=False, server_default='',
index=True)
_password = db.Column(db.LargeBinary, name="password",
nullable=False)
note = db.Column(db.String(255), nullable=True)
settings = db.Column(db.MutableDict.as_mutable(db.MarshalBinary(
default_value=lambda: dict(), force_type=dict)),
nullable=True)
nickname = db.Column(db.String(255), nullable=False, server_default='',
index=True)
last_login = db.Column(db.DateTime, nullable=True)
All your models used in a module have to be located in models.py
inside
your module package (see Modules for developers).
Note
If you have any relations between tables use ForeignKey
definitions and SQLAlchemy Relationships.
Quering¶
Once you have written models it is possible to change the old way of writing SQL queries:
>>> run_sql("SELECT user.nickname FROM user")
[u.nickname for u in User.query.all()]
# Old Invenio way ...
>>> User.query.values(User.nickname)
>>> db.session.query(User.nickname).all()
>>> db.select([User.nickname]).execute().fetchall()
# All roads lead to Rome ... however some are slower.
We also need a WHERE clause in our SQL statements. Let’s prepare a statement for a list with all messages sent by a user:
>>> db.select([User.nickname, MsgMESSAGE.subject]).execute().fetchall()
[('admin', 'test1'),
('admin', 'test2'),
('jekyll', 'test1'),
('jekyll', 'test2'),
...
('balthasar', 'test1'),
('balthasar', 'test2')]
# Something is missing in our query ...
>>> db.select([User.nickname, MsgMESSAGE.subject], User.id==MsgMESSAGE.id_user_from).execute().fetchall()
[('admin', 'test1'),
('admin', 'test2')]
# This is much better.
Operators:
&
,and_
(redefined because of Python operator priorites)|
,or_
~
,not_
==
,<=
,=>
,<
,>
func.like
text operator
+
:>>> str(MsgMESSAGE.subject + ': ' + MsgMESSAGE.body) '"msgMESSAGE".subject || :subject_1 || "msgMESSAGE".body' # :subject_1 will be replaced by ': ' during query execution
Let’s use ORM for getting messages sent by “admin”:
>>> admin = User.query.filter(User.nickname.like('admin')).one()
>>> admin
admin <root@localhost>
>>> admin.sent_messages
[From: admin<root@localhost>, Subject: <test1> body1,
From: admin<root@localhost>, Subject: <test2> body2]
>>> User.query.filter(User.nickname.like('%a%'))
[admin <root@localhost>,
dorian <dorian.gray@cds.cern.ch>,
balthasar <balthasar.montague@cds.cern.ch>
Which brings us to another example where we create ‘’reusable’’ queries using db.bindparam instead of an actual filter value:
>>> q = User.query.filter(User.nickname.like(db.bindparam('nickname')))
>>> q.params({'nickname':'admin'}).one()
admin <root@localhost>
>>> q.params({'nickname':'%a%'}).all()
[admin <root@localhost>,
dorian <dorian.gray@cds.cern.ch>,
balthasar <balthasar.montague@cds.cern.ch>]
Subqueries¶
Let’s start with simple example:
>>> s = db.session.query(User.id).filter(User.nickname.like("%a%")).subquery()
>>> MsgMESSAGE.query.filter(MsgMESSAGE.id_user_from.in_(s)).all()
You can combine subqueries with the delete statement:
>>> sub = db.session.query(UserMsgMESSAGE.id_user_to, UserMsgMESSAGE.id_msgMESSAGE).outerjoin(User, User.id==UserMsgMESSAGE.id_user_to).outerjoin(MsgMESSAGE, UserMsgMESSAGE.id_msgMESSAGE==MsgMESSAGE.id).filter(db.or_(User.id==None, MsgMESSAGE.id==None)).all()
# Find links to not existing messages or users.
>>> db.session.query(UserMsgMESSAGE).filter(db.tuple_(UserMsgMESSAGE.id_user_to, UserMsgMESSAGE.id_msgMESSAGE).in_(sub)).delete(synchronize_session=False)
# Delete messages found in subquery.
Schema¶
When you load all models, you want it to be easy to print create table statements for these models:
>>> for table in db.metadata.tables.values(): print CreateTable(table, on=db.engine.name, bind=db.engine)
Similarly, we can print relevant create statements for indexes:
>>> [str(CreateIndex(i, on=db.engine.name, bind=db.engine)) for i in table.indexes for table in db.metadata.tables.values() if hasattr(table, 'indexes')]
Improve code readability¶
Queries and filters can get quite long and some parts are unnecessarily copied many times.
Some examples follow:
reminder_status = CFG_WEBMESSAGE_STATUS_CODE['REMINDER']
db.session.query(UserMsgMESSAGE).join(User, MsgMESSAGE).filter(db.not_(AsBINARY(UserMsgMESSAGE.status.__eq__(reminder_status))) & (UserMsgMESSAGE.id_user_to == 1)).all()
db.session.query(UserMsgMESSAGE).join(User).filter(
(User.id == MsgMESSAGE.id_user_from) & (UserMsgMESSAGE.id_msgMESSAGE == MsgMESSAGE.id) &
db.not_(AsBINARY(UserMsgMESSAGE.status.__eq__(reminder_status))) & (UserMsgMESSAGE.id_user_to == 1)).all()
filter_all_message_from_user = lambda uid, status: (User.id == MsgMESSAGE.id_user_from) & (UserMsgMESSAGE.id_msgMESSAGE == MsgMESSAGE.id) & db.not_(AsBINARY(UserMsgMESSAGE.status.__eq__(status))) & (UserMsgMESSAGE.id_user_to == uid)
[SQLAlchemy2013] | SQLAlchemy website: http://www.sqlalchemy.org/ |