This package provides SQLAlchemy integration in cone.app
and basic
application nodes for publishing SQLAlchemy models.
Include cone.sql
to install dependencies in your application's
setup.py
.
Adopt your application config ini file to define database location and hook up the related elements to the WSGI pipeline.
[app:my_app]
use = egg:cone.app#main
pyramid.includes =
pyramid_retry
pyramid_tm
tm.commit_veto = pyramid_tm.default_commit_veto
cone.plugins =
cone.sql
sql.db.url = sqlite:///%(here)s/var/sqlite/my_db.db
[filter:remote_addr]
# for use behind nginx
use = egg:cone.app#remote_addr
[filter:session]
use = egg:cone.sql#session
[pipeline:main]
pipeline =
remote_addr
session
my_app
Define the SQLAlchemy model.
from cone.sql import SQLBase
from cone.sql.model import GUID
from sqlalchemy import Column
from sqlalchemy import String
class MyRecord(SQLBase):
__tablename__ = 'my_table'
uid_key = Column(GUID, primary_key=True)
field = Column(String)
Define an application node which represents the SQL row and uses the SQLAlchemy model. The class holds a reference to the related SQLAlchemy model.
from cone.sql.model import SQLRowNode
class MyNode(SQLRowNode):
record_class = MyRecord
Define an application node which represents the table and acts as container for the SQL row nodes. The class holds a reference to the related SQLAlchemy model and the related SQLRowNode.
from cone.sql.model import SQLTableNode
class MyContainer(SQLTableNode):
record_class = MyRecord
child_factory = MyNode
The node name maps to the primary key of the SQLAlchemy model (currenly no
multiple primary keys are supported). Node names are converted to the
primary key data type automatically. The conversion factories are defined at
SQLTableNode.data_type_converters
which can be extended by more data types
if needed.
>>> SQLTableNode.data_type_converters
{<class 'sqlalchemy.sql.sqltypes.String'>: <type 'unicode'>,
<class 'cone.sql.model.GUID'>: <class 'uuid.UUID'>,
<class 'sqlalchemy.sql.sqltypes.Integer'>: <type 'int'>}
In order to publish a SQL table node, the table node must be hooked up to the application model. To hook up the at root level, register it as entry.
import cone.app
cone.app.register_entry('container', MyContainer)
There exists a sql_session_setup
decorator which can be used to perform
session setup tasks like registering SQLAlchemy event listeners.
from cone.sql import sql_session_setup
from sqlalchemy import event
def after_flush(session, flush_context):
"""Do something after flush.
"""
@sql_session_setup
def bind_session_listener(session):
"""SQL session setup callback.
"""
event.listen(session, 'after_flush', after_flush)
Querying the database is done via SQLAlchemy. If you are in a request/response
cycle, you should acquire the session from request via get_session
and
perform arbitrary operations on it. By reading the session from request we ensure
the transaction manager to work properly if configured.
from cone.sql import get_session
session = get_session(request)
result = session.query(MyRecord).all()
If you need a session outside a request/response cycle you can create one by using
the session_factory
.
from cone.sql import session_factory
session = session_factory()
result = session.query(MyRecord).all()
session.close()
SQL based Principal ACL's are implemented in cone.sql.acl
. The related
table gets created as soon as you import from this module.
Using SQLPrincipalACL
requires the model to implement node.interfaces.IUUID
.
from cone.sql.acl import SQLPrincipalACL
from node.base import BaseNode
from node.interfaces import IUUID
from plumber import plumbing
from pyramid.security import Allow
from zope.interface import implementer
import uuid as uuid_module
@implementer(IUUID)
@plumbing(SQLPrincipalACL)
class SQLPrincipalACLNode(BaseNode):
uuid = uuid_module.UUID('1a82fa87-08d6-4e48-8bc2-97ee5a52726d')
@property
def __acl__(self):
return [
(Allow, 'role:editor', ['edit']),
(Allow, 'role:manager', ['manage']),
]
cone.sql.ugm
contains an implementation of the UGM contracts defined at
node.ext.ugm.interfaces
, using sql as backend storage:
+------------+ | Principal | |(data: JSON)| +------------+ ^ | +-----------------------------------------+ | | | | +------+ +-------+ | User | | Group | +------+ +-------+ 1 1 | | | | +-------------+ +-------------+ | | n m | | +-----------------+ | GroupAssignment | +-----------------+
Currently SQLite and PostgreSQL are supported and tested, other DBs must be evaluated concerning their JSON capabilities since users and groups store additional payload data in a JSON field which brings the flexibility to store arbitrary data as a dict in the JSON field.
To activate SQL based UGM backend, it needs to be configured via the application ini config file.:
ugm.backend = sql
sql.user_attrs = id, mail, fullname, portrait
sql.group_attrs = description
sql.binary_attrs = portrait
sql.log_auth = True
sql.user_expires_attr = expires
UGM users and groups are stored in the same database as defined at
sql.db.url
in the config file.
UGM dedicated config options:
sql.user_attrs
is a comma separated list of strings defining the available user attributes stored in the user JSON data field.sql.group_attrs
is a comma separated list of strings defining the available group attributes stored in the group JSON data field.sql.binary_attrs
is a comma separated list of strings defining the attributes which are considered binary and get stored base 64 encoded in the JSON data field of users and groups.sql.log_auth
defaults to False. If set, the first login timestamp will be stored during the first authentication and latest login timestamp will be updated for each successful authentication.sql.user_expires_attr
defaults to None. If set, user expiration is enabled and the value given is the attribute name of the JSON data field where the expiration timestamp gets stored.
Users and groups can be managed with cone.ugm
. If activated,
sql.user_attrs
and sql.group_attrs
can be omitted, relevant information
gets extracted from the ugm.xml
config file.
ugm.backend = sql
ugm.config = %(here)s/ugm.xml
sql.log_auth = True
cone.plugins =
cone.ugm
cone.sql
- Support multiple primary keys.
- Robert Niederreiter (Author)
- Phil Auersperg