mongodb_beaker

MongoDB. backend for Beaker.'s caching / session system.

Based upon Beaker.'s ext:memcache code.

This is implemented in a dont-assume-its-there manner. It uses the beaker namespace as the mongodb row's _id, with everything in that namespace ( e.g. a session or cache namespace) stored as a full document. Each key/value is part of that compound document, using upserts for performance.

I will probably add a toggleable option for using subcollections, as in certain cases such as caching mako templates, this may be desirable / preferred performance wise.

Right now, this is primarily optimized for usage with beaker sessions, although I need to look at tweaking beaker session itself as having it store in individual keys rather than everything in a 'session' key may be desirable for pruning/management/querying.

I have not tackled expiration yet, so you may want to hold off using this if you need it. It will be in the next update, but limits usefulness primarily to sessions right now. (I'll tack a cleanup script in later as well).

Due to the use of upserts, no check-insert is required, but it will overwrite previous values which should be expected behavior while caching. Safe is NOT invoked, so failure will be quiet. TODO - Safe as overridable config option?

Note that, unless you disable. it, the mongodb_beaker container will use pickle (tries loading cpickle first, falls back on pickle) to serialize/deserialize data to MongoDB.

Beaker should maintain thread safety on connections internally and so I am relying upon that rather than setting up threadlocal, etc. If this assumption is wrong or you run into issues, please let me know.

Configuration

To set this up in your own project so that beaker can find it, it must define a setuptools entry point in your setup.py file. If you install from the egg distribution, mongodb_beaker's setup.py SHOULD create a beaker.backend entry point. If you need to tweak it/see how it's done or it just doesn't work and you need to define your own, mine looks like this:

>>> entry_points="""
... [beaker.backends]
... mongodb = mongodb_beaker:MongoDBNamespaceManager
... """,

With this defined, beaker should automatically find the entry point at startup (Beaker 1.4 and higher support custom entry points) and load it as an optional backend called 'mongodb'. There are several ways to configure Beaker, I only cover ini file (such as with Pylons) here. There are more configuration options and details in the Beaker configuration docs¹.

I have a few cache regions in one of my applications, some of which are memcache and some are on mongodb. The region config looks like this:

>>> # new style cache settings
... beaker.cache.regions = comic_archives, navigation
... beaker.cache.comic_archives.type = libmemcached
... beaker.cache.comic_archives.url = 127.0.0.1:11211
... beaker.cache.comic_archives.expire = 604800
... beaker.cache.navigation.type = mongodb
... beaker.cache.navigation.url = mongodb://localhost:27017/beaker.navigation
... beaker.cache.navigation.expire = 86400

The Beaker docs[1] contain detailed information on configuring regions. The item we're interested in here is the beaker.cache.navigation keys. Each beaker cache definition needs a type field, which defines which backend to use. Specifying mongodb will (if the module is properly installed) tell Beaker to cache via mongodb. Note that if Beaker cannot load the extension, it will tell you that mongodb is an invalid backend.

Expiration is standard beaker syntax, although not supported at the moment in this backend.

Finally, you need to define a URL to connect to MongoDB. This follows the standardized MongoDB URI Format². Currently the only options supported is 'slaveOK'. For backwards compatibility with old versions of mongodb_beaker, separating database and collection with a '#' instead of '.' is supported, but deprecated. The syntax is mongodb://<hostname>[:port]/<database>.<collection>

You must define a collection for MongoDB to store data in, in addition to a database.

If you want to use MongoDB's optional authentication support, that is also supported. Simply define your URL as such:

>>> beaker.cache.navigation.url = mongodb://bwmcadams@passW0Rd?@localhost:27017/beaker.navigation

The mongodb_beaker backend will attempt to authenticate with the username and password. You must configure MongoDB's optional authentication support³ for this to work (By default MongoDB doesn't use authentication).

Reading from Secondaries (SlaveOK)

If you'd like to enable reading from secondaries (SlaveOK), you can add that to your URL:

>>> beaker.cache.navigation.url = mongodb://bwmcadams@passW0Rd?@localhost:27017/beaker.navigation?slaveok=true

Using Beaker Sessions and disabling pickling

If you want to save some CPU cycles and can guarantee that what you're passing in is either "mongo-safe" and doesn't need pickling, or you know it's already pickled (such as while using beaker sessions), you can set an extra beaker config flag of skip_pickle=True. .. admonition:: To make that perfectly clear, Beaker sessions are ALREADY PASSED IN pickled, so you want to configure it to skip_pickle. It shouldn't hurt anything to double-pickle, but you will certainly waste precious CPU cycles. And wasting CPU cycles is kind of counterproductive in a caching system.

My pylons application configuration for mongodb_beaker has the following session_configuration:

>>> beaker.session.type = mongodb
... beaker.session.url = mongodb://localhost:27017/beaker.sessions
... beaker.session.skip_pickle = True

Depending on your individual needs, you may also wish to create a capped collection for your caching (e.g. memcache-like only most recently used storage)

See the MongoDB CappedCollection. docs for details.

Sparse Collection Support

The default behavior of mongodb_beaker is to create a single MongoDB Document for each namespace, and store each cache key/value on that document. In this case, the "_id" of the document will be the namespace, and each new cache entry will be attached to that document.

This approach works well in many cases and makes it very easy for Mongo to efficiently manage your cache. However, in other cases you may wish to change behavior. This may be for efficiency reasons, or because you're worried about documents getting too large.

In this case, you can enable a "sparse collection" mode, where mongodb_beaker will create a document for EACH key in the namespace. When sparse collections are enabled, the "_id" of a document is a compound document containing the namespace and the key:

{ "_id" : { "namespace" : "testcache", "key" : "value" } }

The cache data for that key will be stored in a document field 'data'. You can enable sparse collections in your config with the 'sparse_collections' variable:

>>> beaker.session.type = mongodb
... beaker.session.url = mongodb://localhost:27017/beaker.sessions
... beaker.session.sparse_collections = True

Note for Users of Previous Releases

For bug fix and feature reasons, MongoDB Beaker 0.5+ are not compatible with caches created by previous releases. Because this is cache data, it shouldn't be a big deal. We recommend dropping or flushing your entire cache collection(s) before upgrading to 0.5+ and be aware that it will generate new caches.