zope.session 4.2.0

Design Issues

Sessions introduce a number of issues to be considered:

• Clients have to be identified. A number of approaches are possible, including:

  o Using HTTP cookies. The application assigns a client identifier,

  which is stored in a cookie. This technique is the most straightforward, but can be defeated if the client does not support HTTP cookies (usually because the feature has been disabled).

  o Using URLs. The application assigns a client identifier, which is

  stored in the URL. This makes URLs a bit uglier and requires some care. If people copy URLs and send them to others, then you could end up with multiple clients with the same session identifier. There are a number of ways to reduce the risk of accidental reuse of session identifiers:

  • Embed the client IP address in the identifier

  • Expire the identifier

  o Use hidden form variables. This complicates applications. It

  requires all requests to be POST requests and requires the maintenance of the hidden variables.

  o Use the client IP address

  This doesn’t work very well, because an IP address may be shared by many clients.

• Data storage

  Data can be simply stored in the object database. This provides lots of flexibility. You can store pretty much anything you want as long as it is persistent. You get the full benefit of the object database, such as transactions, transparency, clustering, and so on. Using the object database is especially useful when:

  • Writes are infrequent

  • Data are complex

  If writes are frequent, then the object database introduces scalability problems. Really, any transactional database is likely to introduce problems with frequent writes. If you are tempted to update session data on every request, think very hard about it. You are creating a scalability problem.

  If you know that scalability is not (and never will be) an issue, you can just use the object database.

  If you have client data that needs to be updated often (as in every request), consider storing the data on the client. (Like all data received from a client, it may be tainted and, in most instances, should not be trusted. Sensitive information that the user should not see should likewise not be stored on the client, unless encrypted with a key the client has no access to.) If you can’t store it on the client, then consider some other storage mechanism, like a fast database, possibly without transaction support.

  You may be tempted to store session data in memory for speed. This doesn’t turn out to work very well. If you need scalability, then you need to be able to use an application-server cluster and storage of session data in memory defeats that. You can use “server-affinity” to assure that requests from a client always go back to the same server, but not all load balancers support server affinity, and, for those that do, enabling server affinity tends to defeat load balancing.

• Session expiration

  You may wish to ensure that sessions terminate after some period of time. This may be for security reasons, or to avoid accidental sharing of a session among multiple clients. The policy might be expressed in terms of total session time, or maximum inactive time, or some combination.

  There are a number of ways to approach this. You can expire client ids. You can expire session data.

• Data expiration

  Because HTTP is a stateless protocol, you can’t tell whether a user is thinking about a task or has simply stopped working on it. Some means is needed to free server session storage that is no-longer needed.

  The simplest strategy is to never remove data. This strategy has some obvious disadvantages. Other strategies can be viewed as optimizations of the basic strategy. It is important to realize that a data expiration strategy can be informed by, but need not be constrained by a session-expiration strategy.

Overview

Sessions allow us to fake state over a stateless protocol - HTTP. We do this by having a unique identifier stored across multiple HTTP requests, be it a cookie or some id mangled into the URL.

The IClientIdManager Utility provides this unique id. It is responsible for propagating this id so that future requests from the client get the same id (eg. by setting an HTTP cookie). This utility is used when we adapt the request to the unique client id:

>>> from zope.session.interfaces import IClientId
>>> client_id = IClientId(request)

The ISession adapter gives us a mapping that can be used to store and retrieve session data. A unique key (the package id) is used to avoid namespace clashes:

>>> from zope.session.interfaces import ISession
>>> pkg_id = 'products.foo'
>>> session = ISession(request)[pkg_id]
>>> session['color'] = 'red'

>>> session2 = ISession(request)['products.bar']
>>> session2['color'] = 'blue'

>>> session['color']
'red'
>>> session2['color']
'blue'

Data Storage

The actual data is stored in an ISessionDataContainer utility. ISession chooses which ISessionDataContainer should be used by looking up as a named utility using the package id. This allows the site administrator to configure where the session data is actually stored by adding a registration for desired ISessionDataContainer with the correct name.

>>> import zope.component
>>> from zope.session.interfaces import ISessionDataContainer
>>> sdc = zope.component.getUtility(ISessionDataContainer, pkg_id)
>>> sdc[client_id][pkg_id] is session
True
>>> sdc[client_id][pkg_id]['color']
'red'

If no ISessionDataContainer utility can be located by name using the package id, then the unnamed ISessionDataContainer utility is used as a fallback. An unnamed ISessionDataContainer is automatically created for you, which may replaced with a different implementation if desired.

>>> ISession(request)['unknown'] \
...     is zope.component.getUtility(ISessionDataContainer)[client_id]\
...         ['unknown']
True

The ISessionDataContainer contains ISessionData objects, and ISessionData objects in turn contain ISessionPkgData objects. You should never need to know this unless you are writing administrative views for the session machinery.

>>> from zope.session.interfaces import ISessionData
>>> from zope.session.interfaces import ISessionPkgData
>>> ISessionData.providedBy(sdc[client_id])
True
>>> ISessionPkgData.providedBy(sdc[client_id][pkg_id])
True

The ISessionDataContainer is responsible for expiring session data. The expiry time can be configured by settings its timeout attribute.

>>> sdc.timeout = 1200 # 1200 seconds or 20 minutes

Restrictions

Data stored in the session must be persistent or picklable. (Exactly which builtin and standard objects can be pickled depends on the Python version, the Python implementation, and the ZODB version, so we demonstrate with a custom object.)

>>> import transaction
>>> class NoPickle(object):
...     def __reduce__(self):
...          raise TypeError("I cannot be pickled")
>>> session['oops'] = NoPickle()
>>> transaction.commit()
Traceback (most recent call last):
    [...]
TypeError: I cannot be pickled

Clean up:

>>> transaction.abort()

Page Templates

Session data may be accessed in page template documents using TALES:

<span tal:content="request/session:products.foo/color | default">
    green
</span>

or:

<div tal:define="session request/session:products.foo">
    <script type="text/server-python">
        try:
            session['count'] += 1
        except KeyError:
            session['count'] = 1
    </script>

    <span tal:content="session/count" />
</div>

Session Timeout

Sessions have a timeout (defaulting to an hour, in seconds).

>>> import zope.session.session
>>> data_container = zope.session.session.PersistentSessionDataContainer()
>>> data_container.timeout
3600

We need to keep up with when the session was last used (to know when it needs to be expired), but it would be too resource-intensive to write the last access time every, single time the session data is touched. The session machinery compromises by only recording the last access time periodically. That period is called the “resolution”. That also means that if the last-access-time + the-resolution < now, then the session is considered to have timed out.

The default resolution is 10 minutes (600 seconds), meaning that a users session will actually time out sometime between 50 and 60 minutes.

>>> data_container.resolution
600